Release: v0.30.2

update to a place holder

README.md

@@ -202,7 +202,6 @@ Also, say 👋 in our public Discord channel <a href="https://discord.gg/G7tWnz9
 
 - https://github.com/microsoft/TaskMatrix
 - https://github.com/invoke-ai/InvokeAI
-- https://github.com/InstantID/InstantID
 - https://github.com/apple/ml-stable-diffusion
 - https://github.com/Sanster/lama-cleaner
 - https://github.com/IDEA-Research/Grounded-Segment-Anything

docs/source/en/_toctree.yml

@@ -223,76 +223,68 @@
     sections:
     - local: api/models/overview
       title: Overview
-    - sections:
-      - local: api/models/controlnet
-        title: ControlNetModel
-      - local: api/models/controlnet_hunyuandit
-        title: HunyuanDiT2DControlNetModel
-      - local: api/models/controlnet_sd3
-        title: SD3ControlNetModel
-      - local: api/models/controlnet_sparsectrl
-        title: SparseControlNetModel
-      title: ControlNets
-    - sections:
-      - local: api/models/aura_flow_transformer2d
-        title: AuraFlowTransformer2DModel
-      - local: api/models/cogvideox_transformer3d
-        title: CogVideoXTransformer3DModel
-      - local: api/models/dit_transformer2d
-        title: DiTTransformer2DModel
-      - local: api/models/flux_transformer
-        title: FluxTransformer2DModel
-      - local: api/models/hunyuan_transformer2d
-        title: HunyuanDiT2DModel
-      - local: api/models/latte_transformer3d
-        title: LatteTransformer3DModel
-      - local: api/models/lumina_nextdit2d
-        title: LuminaNextDiT2DModel
-      - local: api/models/pixart_transformer2d
-        title: PixArtTransformer2DModel
-      - local: api/models/prior_transformer
-        title: PriorTransformer
-      - local: api/models/sd3_transformer2d
-        title: SD3Transformer2DModel
-      - local: api/models/stable_audio_transformer
-        title: StableAudioDiTModel
-      - local: api/models/transformer2d
-        title: Transformer2DModel
-      - local: api/models/transformer_temporal
-        title: TransformerTemporalModel
-      title: Transformers
-    - sections:
-      - local: api/models/stable_cascade_unet
-        title: StableCascadeUNet
-      - local: api/models/unet
-        title: UNet1DModel
-      - local: api/models/unet2d
-        title: UNet2DModel
-      - local: api/models/unet2d-cond
-        title: UNet2DConditionModel
-      - local: api/models/unet3d-cond
-        title: UNet3DConditionModel
-      - local: api/models/unet-motion
-        title: UNetMotionModel
-      - local: api/models/uvit2d
-        title: UViT2DModel
-      title: UNets
-    - sections:
-      - local: api/models/autoencoderkl
-        title: AutoencoderKL
-      - local: api/models/autoencoderkl_cogvideox
-        title: AutoencoderKLCogVideoX
-      - local: api/models/asymmetricautoencoderkl
-        title: AsymmetricAutoencoderKL
-      - local: api/models/consistency_decoder_vae
-        title: ConsistencyDecoderVAE
-      - local: api/models/autoencoder_oobleck
-        title: Oobleck AutoEncoder
-      - local: api/models/autoencoder_tiny
-        title: Tiny AutoEncoder
-      - local: api/models/vq
-        title: VQModel
-      title: VAEs
+    - local: api/models/unet
+      title: UNet1DModel
+    - local: api/models/unet2d
+      title: UNet2DModel
+    - local: api/models/unet2d-cond
+      title: UNet2DConditionModel
+    - local: api/models/unet3d-cond
+      title: UNet3DConditionModel
+    - local: api/models/unet-motion
+      title: UNetMotionModel
+    - local: api/models/uvit2d
+      title: UViT2DModel
+    - local: api/models/vq
+      title: VQModel
+    - local: api/models/autoencoderkl
+      title: AutoencoderKL
+    - local: api/models/autoencoderkl_cogvideox
+      title: AutoencoderKLCogVideoX
+    - local: api/models/asymmetricautoencoderkl
+      title: AsymmetricAutoencoderKL
+    - local: api/models/stable_cascade_unet
+      title: StableCascadeUNet
+    - local: api/models/autoencoder_tiny
+      title: Tiny AutoEncoder
+    - local: api/models/autoencoder_oobleck
+      title: Oobleck AutoEncoder
+    - local: api/models/consistency_decoder_vae
+      title: ConsistencyDecoderVAE
+    - local: api/models/transformer2d
+      title: Transformer2DModel
+    - local: api/models/pixart_transformer2d
+      title: PixArtTransformer2DModel
+    - local: api/models/dit_transformer2d
+      title: DiTTransformer2DModel
+    - local: api/models/hunyuan_transformer2d
+      title: HunyuanDiT2DModel
+    - local: api/models/aura_flow_transformer2d
+      title: AuraFlowTransformer2DModel
+    - local: api/models/flux_transformer
+      title: FluxTransformer2DModel
+    - local: api/models/latte_transformer3d
+      title: LatteTransformer3DModel
+    - local: api/models/cogvideox_transformer3d
+      title: CogVideoXTransformer3DModel
+    - local: api/models/lumina_nextdit2d
+      title: LuminaNextDiT2DModel
+    - local: api/models/transformer_temporal
+      title: TransformerTemporalModel
+    - local: api/models/sd3_transformer2d
+      title: SD3Transformer2DModel
+    - local: api/models/stable_audio_transformer
+      title: StableAudioDiTModel
+    - local: api/models/prior_transformer
+      title: PriorTransformer
+    - local: api/models/controlnet
+      title: ControlNetModel
+    - local: api/models/controlnet_hunyuandit
+      title: HunyuanDiT2DControlNetModel
+    - local: api/models/controlnet_sd3
+      title: SD3ControlNetModel
+    - local: api/models/controlnet_sparsectrl
+      title: SparseControlNetModel
     title: Models
   - isExpanded: false
     sections:

docs/source/en/api/pipelines/cogvideox.md

@@ -29,6 +29,10 @@ Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.m
 
 This pipeline was contributed by [zRzRzRzRzRzRzR](https://github.com/zRzRzRzRzRzRzR). The original codebase can be found [here](https://huggingface.co/THUDM). The original weights can be found under [hf.co/THUDM](https://huggingface.co/THUDM).
 
+There are two models available that can be used with the CogVideoX pipeline:
+- [`THUDM/CogVideoX-2b`](https://huggingface.co/THUDM/CogVideoX-2b)
+- [`THUDM/CogVideoX-5b`](https://huggingface.co/THUDM/CogVideoX-5b)
+
 ## Inference
 
 Use [`torch.compile`](https://huggingface.co/docs/diffusers/main/en/tutorials/fast_diffusion#torchcompile) to reduce the inference latency.
@@ -68,7 +72,7 @@ With torch.compile(): Average inference time: 76.27 seconds.
 
 ### Memory optimization
 
-CogVideoX requires about 19 GB of GPU memory to decode 49 frames (6 seconds of video at 8 FPS) with output resolution 720x480 (W x H), which makes it not possible to run on consumer GPUs or free-tier T4 Colab. The following memory optimizations could be used to reduce the memory footprint. For replication, you can refer to [this](https://gist.github.com/a-r-r-o-w/3959a03f15be5c9bd1fe545b09dfcc93) script.
+CogVideoX-2b requires about 19 GB of GPU memory to decode 49 frames (6 seconds of video at 8 FPS) with output resolution 720x480 (W x H), which makes it not possible to run on consumer GPUs or free-tier T4 Colab. The following memory optimizations could be used to reduce the memory footprint. For replication, you can refer to [this](https://gist.github.com/a-r-r-o-w/3959a03f15be5c9bd1fe545b09dfcc93) script.
 
 - `pipe.enable_model_cpu_offload()`:
   - Without enabling cpu offloading, memory usage is `33 GB`

docs/source/en/api/pipelines/controlnet_sd3.md

@@ -1,4 +1,4 @@
-<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+<!--Copyright 2023 The HuggingFace Team and The InstantX Team. All rights reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
 the License. You may obtain a copy of the License at
@@ -22,16 +22,7 @@ The abstract from the paper is:
 
 *We present ControlNet, a neural network architecture to add spatial conditioning controls to large, pretrained text-to-image diffusion models. ControlNet locks the production-ready large diffusion models, and reuses their deep and robust encoding layers pretrained with billions of images as a strong backbone to learn a diverse set of conditional controls. The neural architecture is connected with "zero convolutions" (zero-initialized convolution layers) that progressively grow the parameters from zero and ensure that no harmful noise could affect the finetuning. We test various conditioning controls, eg, edges, depth, segmentation, human pose, etc, with Stable Diffusion, using single or multiple conditions, with or without prompts. We show that the training of ControlNets is robust with small (<50k) and large (>1m) datasets. Extensive results show that ControlNet may facilitate wider applications to control image diffusion models.*
 
-This controlnet code is mainly implemented by [The InstantX Team](https://huggingface.co/InstantX). The inpainting-related code was developed by [The Alimama Creative Team](https://huggingface.co/alimama-creative). You can find pre-trained checkpoints for SD3-ControlNet in the table below:
-
-
-| ControlNet type | Developer | Link |
-| -------- | ---------- | ---- |
-| Canny | [The InstantX Team](https://huggingface.co/InstantX) | [Link](https://huggingface.co/InstantX/SD3-Controlnet-Canny) |
-| Pose | [The InstantX Team](https://huggingface.co/InstantX) | [Link](https://huggingface.co/InstantX/SD3-Controlnet-Pose) |
-| Tile | [The InstantX Team](https://huggingface.co/InstantX) | [Link](https://huggingface.co/InstantX/SD3-Controlnet-Tile) |
-| Inpainting | [The AlimamaCreative Team](https://huggingface.co/alimama-creative) | [link](https://huggingface.co/alimama-creative/SD3-Controlnet-Inpainting) |
-
+This code is implemented by [The InstantX Team](https://huggingface.co/InstantX). You can find pre-trained checkpoints for SD3-ControlNet on [The InstantX Team](https://huggingface.co/InstantX) Hub profile.
 
 <Tip>
 
@@ -44,10 +35,5 @@ Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers)
 	- all
 	- __call__
 
-## StableDiffusion3ControlNetInpaintingPipeline
-[[autodoc]] pipelines.controlnet_sd3.pipeline_stable_diffusion_3_controlnet_inpainting.StableDiffusion3ControlNetInpaintingPipeline
-	- all
-	- __call__
-
 ## StableDiffusion3PipelineOutput
 [[autodoc]] pipelines.stable_diffusion_3.pipeline_output.StableDiffusion3PipelineOutput

docs/source/en/api/pipelines/kolors.md

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 ![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/kolors/kolors_header_collage.png)
 
-Kolors is a large-scale text-to-image generation model based on latent diffusion, developed by [the Kuaishou Kolors team](https://github.com/Kwai-Kolors/Kolors). Trained on billions of text-image pairs, Kolors exhibits significant advantages over both open-source and closed-source models in visual quality, complex semantic accuracy, and text rendering for both Chinese and English characters. Furthermore, Kolors supports both Chinese and English inputs, demonstrating strong performance in understanding and generating Chinese-specific content. For more details, please refer to this [technical report](https://github.com/Kwai-Kolors/Kolors/blob/master/imgs/Kolors_paper.pdf).
+Kolors is a large-scale text-to-image generation model based on latent diffusion, developed by [the Kuaishou Kolors team](kwai-kolors@kuaishou.com). Trained on billions of text-image pairs, Kolors exhibits significant advantages over both open-source and closed-source models in visual quality, complex semantic accuracy, and text rendering for both Chinese and English characters. Furthermore, Kolors supports both Chinese and English inputs, demonstrating strong performance in understanding and generating Chinese-specific content. For more details, please refer to this [technical report](https://github.com/Kwai-Kolors/Kolors/blob/master/imgs/Kolors_paper.pdf).
 
 The abstract from the technical report is:
 
@@ -74,7 +74,7 @@ image_encoder = CLIPVisionModelWithProjection.from_pretrained(
 
 pipe = KolorsPipeline.from_pretrained(
     "Kwai-Kolors/Kolors-diffusers", image_encoder=image_encoder, torch_dtype=torch.float16, variant="fp16"
-)
+).to("cuda")
 pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config, use_karras_sigmas=True)
 
 pipe.load_ip_adapter(

docs/source/en/api/pipelines/pag.md

@@ -20,7 +20,7 @@ The abstract from the paper is:
 
 *Recent studies have demonstrated that diffusion models are capable of generating high-quality samples, but their quality heavily depends on sampling guidance techniques, such as classifier guidance (CG) and classifier-free guidance (CFG). These techniques are often not applicable in unconditional generation or in various downstream tasks such as image restoration. In this paper, we propose a novel sampling guidance, called Perturbed-Attention Guidance (PAG), which improves diffusion sample quality across both unconditional and conditional settings, achieving this without requiring additional training or the integration of external modules. PAG is designed to progressively enhance the structure of samples throughout the denoising process. It involves generating intermediate samples with degraded structure by substituting selected self-attention maps in diffusion U-Net with an identity matrix, by considering the self-attention mechanisms' ability to capture structural information, and guiding the denoising process away from these degraded samples. In both ADM and Stable Diffusion, PAG surprisingly improves sample quality in conditional and even unconditional scenarios. Moreover, PAG significantly improves the baseline performance in various downstream tasks where existing guidances such as CG or CFG cannot be fully utilized, including ControlNet with empty prompts and image restoration such as inpainting and deblurring.*
 
-PAG can be used by specifying the `pag_applied_layers` as a parameter when instantiating a PAG pipeline. It can be a single string or a list of strings. Each string can be a unique layer identifier or a regular expression to identify one or more layers.
+PAG can be used by specifying the `pag_applied_layers` as a parameter when instantiating a PAG pipeline. It can be a single string or a list of strings. Each string can be a unique layer identifier or a regular expression to identify one or more layers. 
 
 - Full identifier as a normal string: `down_blocks.2.attentions.0.transformer_blocks.0.attn1.processor`
 - Full identifier as a RegEx: `down_blocks.2.(attentions|motion_modules).0.transformer_blocks.0.attn1.processor`
@@ -46,7 +46,7 @@ Since RegEx is supported as a way for matching layer identifiers, it is crucial
 ## KolorsPAGPipeline
 [[autodoc]] KolorsPAGPipeline
   - all
-  - __call__
+  - __call__  
 
 ## StableDiffusionPAGPipeline
 [[autodoc]] StableDiffusionPAGPipeline

docs/source/en/api/pipelines/stable_audio.md

@@ -21,7 +21,7 @@ Stable Audio is trained on a corpus of around 48k audio recordings, where around
 The abstract of the paper is the following:
 *Open generative models are vitally important for the community, allowing for fine-tunes and serving as baselines when presenting new models. However, most current text-to-audio models are private and not accessible for artists and researchers to build upon. Here we describe the architecture and training process of a new open-weights text-to-audio model trained with Creative Commons data. Our evaluation shows that the model's performance is competitive with the state-of-the-art across various metrics. Notably, the reported FDopenl3 results (measuring the realism of the generations) showcase its potential for high-quality stereo sound synthesis at 44.1kHz.*
 
-This pipeline was contributed by [Yoach Lacombe](https://huggingface.co/ylacombe). The original codebase can be found at [Stability-AI/stable-audio-tools](https://github.com/Stability-AI/stable-audio-tools).
+This pipeline was contributed by [Yoach Lacombe](https://huggingface.co/ylacombe). The original codebase can be found at [Stability-AI/stable-audio-tool](https://github.com/Stability-AI/stable-audio-tool).
 
 ## Tips
 

docs/source/en/optimization/fp16.md

@@ -125,5 +125,3 @@ image
     <figcaption class="mt-2 text-center text-sm text-gray-500">distilled Stable Diffusion + Tiny AutoEncoder</figcaption>
   </div>
 </div>
-
-More tiny autoencoder models for other Stable Diffusion models, like Stable Diffusion 3, are available from [madebyollin](https://huggingface.co/madebyollin).

docs/source/en/training/distributed_inference.md

@@ -48,7 +48,7 @@ accelerate launch run_distributed.py --num_processes=2
 
 <Tip>
 
-Refer to this minimal example [script](https://gist.github.com/sayakpaul/cfaebd221820d7b43fae638b4dfa01ba) for running inference across multiple GPUs. To learn more, take a look at the [Distributed Inference with 🤗 Accelerate](https://huggingface.co/docs/accelerate/en/usage_guides/distributed_inference#distributed-inference-with-accelerate) guide.
+To learn more, take a look at the [Distributed Inference with 🤗 Accelerate](https://huggingface.co/docs/accelerate/en/usage_guides/distributed_inference#distributed-inference-with-accelerate) guide.
 
 </Tip>
 
@@ -108,4 +108,4 @@ torchrun run_distributed.py --nproc_per_node=2
 ```
 
 > [!TIP]
-> You can use `device_map` within a [`DiffusionPipeline`] to distribute its model-level components on multiple devices. Refer to the [Device placement](../tutorials/inference_with_big_models#device-placement) guide to learn more.
+> You can use `device_map` within a [`DiffusionPipeline`] to distribute its model-level components on multiple devices. Refer to the [Device placement](../tutorials/inference_with_big_models#device-placement) guide to learn more.

docs/source/en/using-diffusers/merge_loras.md

@@ -14,9 +14,9 @@ specific language governing permissions and limitations under the License.
 
 It can be fun and creative to use multiple [LoRAs]((https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora)) together to generate something entirely new and unique. This works by merging multiple LoRA weights together to produce images that are a blend of different styles. Diffusers provides a few methods to merge LoRAs depending on *how* you want to merge their weights, which can affect image quality.
 
-This guide will show you how to merge LoRAs using the [`~loaders.PeftAdapterMixin.set_adapters`] and [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) methods. To improve inference speed and reduce memory-usage of merged LoRAs, you'll also see how to use the [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method to fuse the LoRA weights with the original weights of the underlying model.
+This guide will show you how to merge LoRAs using the [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] and [`~peft.LoraModel.add_weighted_adapter`] methods. To improve inference speed and reduce memory-usage of merged LoRAs, you'll also see how to use the [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method to fuse the LoRA weights with the original weights of the underlying model.
 
-For this guide, load a Stable Diffusion XL (SDXL) checkpoint and the [KappaNeuro/studio-ghibli-style](https://huggingface.co/KappaNeuro/studio-ghibli-style) and [Norod78/sdxl-chalkboarddrawing-lora](https://huggingface.co/Norod78/sdxl-chalkboarddrawing-lora) LoRAs with the [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] method. You'll need to assign each LoRA an `adapter_name` to combine them later.
+For this guide, load a Stable Diffusion XL (SDXL) checkpoint and the [KappaNeuro/studio-ghibli-style]() and [Norod78/sdxl-chalkboarddrawing-lora]() LoRAs with the [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] method. You'll need to assign each LoRA an `adapter_name` to combine them later.
 
 ```py
 from diffusers import DiffusionPipeline
@@ -29,7 +29,7 @@ pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_
 
 ## set_adapters
 
-The [`~loaders.PeftAdapterMixin.set_adapters`] method merges LoRA adapters by concatenating their weighted matrices. Use the adapter name to specify which LoRAs to merge, and the `adapter_weights` parameter to control the scaling for each LoRA. For example, if `adapter_weights=[0.5, 0.5]`, then the merged LoRA output is an average of both LoRAs. Try adjusting the adapter weights to see how it affects the generated image!
+The [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] method merges LoRA adapters by concatenating their weighted matrices. Use the adapter name to specify which LoRAs to merge, and the `adapter_weights` parameter to control the scaling for each LoRA. For example, if `adapter_weights=[0.5, 0.5]`, then the merged LoRA output is an average of both LoRAs. Try adjusting the adapter weights to see how it affects the generated image!
 
 ```py
 pipeline.set_adapters(["ikea", "feng"], adapter_weights=[0.7, 0.8])
@@ -47,19 +47,19 @@ image
 ## add_weighted_adapter
 
 > [!WARNING]
-> This is an experimental method that adds PEFTs [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) method to Diffusers to enable more efficient merging methods. Check out this [issue](https://github.com/huggingface/diffusers/issues/6892) if you're interested in learning more about the motivation and design behind this integration.
+> This is an experimental method that adds PEFTs [`~peft.LoraModel.add_weighted_adapter`] method to Diffusers to enable more efficient merging methods. Check out this [issue](https://github.com/huggingface/diffusers/issues/6892) if you're interested in learning more about the motivation and design behind this integration.
 
-The [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) method provides access to more efficient merging method such as [TIES and DARE](https://huggingface.co/docs/peft/developer_guides/model_merging). To use these merging methods, make sure you have the latest stable version of Diffusers and PEFT installed.
+The [`~peft.LoraModel.add_weighted_adapter`] method provides access to more efficient merging method such as [TIES and DARE](https://huggingface.co/docs/peft/developer_guides/model_merging). To use these merging methods, make sure you have the latest stable version of Diffusers and PEFT installed.
 
 ```bash
 pip install -U diffusers peft
 ```
 
-There are three steps to merge LoRAs with the [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) method:
+There are three steps to merge LoRAs with the [`~peft.LoraModel.add_weighted_adapter`] method:
 
-1. Create a [PeftModel](https://huggingface.co/docs/peft/package_reference/peft_model#peft.PeftModel) from the underlying model and LoRA checkpoint.
+1. Create a [`~peft.PeftModel`] from the underlying model and LoRA checkpoint.
 2. Load a base UNet model and the LoRA adapters.
-3. Merge the adapters using the [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) method and the merging method of your choice.
+3. Merge the adapters using the [`~peft.LoraModel.add_weighted_adapter`] method and the merging method of your choice.
 
 Let's dive deeper into what these steps entail.
 
@@ -92,7 +92,7 @@ pipeline = DiffusionPipeline.from_pretrained(
 pipeline.load_lora_weights("ostris/ikea-instructions-lora-sdxl", weight_name="ikea_instructions_xl_v1_5.safetensors", adapter_name="ikea")
 ```
 
-Now you'll create a [PeftModel](https://huggingface.co/docs/peft/package_reference/peft_model#peft.PeftModel) from the loaded LoRA checkpoint by combining the SDXL UNet and the LoRA UNet from the pipeline.
+Now you'll create a [`~peft.PeftModel`] from the loaded LoRA checkpoint by combining the SDXL UNet and the LoRA UNet from the pipeline.
 
 ```python
 from peft import get_peft_model, LoraConfig
@@ -112,7 +112,7 @@ ikea_peft_model.load_state_dict(original_state_dict, strict=True)
 > [!TIP]
 > You can optionally push the ikea_peft_model to the Hub by calling `ikea_peft_model.push_to_hub("ikea_peft_model", token=TOKEN)`.
 
-Repeat this process to create a [PeftModel](https://huggingface.co/docs/peft/package_reference/peft_model#peft.PeftModel) from the [lordjia/by-feng-zikai](https://huggingface.co/lordjia/by-feng-zikai) LoRA.
+Repeat this process to create a [`~peft.PeftModel`] from the [lordjia/by-feng-zikai](https://huggingface.co/lordjia/by-feng-zikai) LoRA.
 
 ```python
 pipeline.delete_adapters("ikea")
@@ -148,7 +148,7 @@ model = PeftModel.from_pretrained(base_unet, "stevhliu/ikea_peft_model", use_saf
 model.load_adapter("stevhliu/feng_peft_model", use_safetensors=True, subfolder="feng", adapter_name="feng")
 ```
 
-3. Merge the adapters using the [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) method and the merging method of your choice (learn more about other merging methods in this [blog post](https://huggingface.co/blog/peft_merging)). For this example, let's use the `"dare_linear"` method to merge the LoRAs.
+3. Merge the adapters using the [`~peft.LoraModel.add_weighted_adapter`] method and the merging method of your choice (learn more about other merging methods in this [blog post](https://huggingface.co/blog/peft_merging)). For this example, let's use the `"dare_linear"` method to merge the LoRAs.
 
 > [!WARNING]
 > Keep in mind the LoRAs need to have the same rank to be merged!
@@ -182,9 +182,9 @@ image
 
 ## fuse_lora
 
-Both the [`~loaders.PeftAdapterMixin.set_adapters`] and [add_weighted_adapter](https://huggingface.co/docs/peft/package_reference/lora#peft.LoraModel.add_weighted_adapter) methods require loading the base model and the LoRA adapters separately which incurs some overhead. The [`~loaders.lora_base.LoraBaseMixin.fuse_lora`] method allows you to fuse the LoRA weights directly with the original weights of the underlying model. This way, you're only loading the model once which can increase inference and lower memory-usage.
+Both the [`~loaders.UNet2DConditionLoadersMixin.set_adapters`] and [`~peft.LoraModel.add_weighted_adapter`] methods require loading the base model and the LoRA adapters separately which incurs some overhead. The [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method allows you to fuse the LoRA weights directly with the original weights of the underlying model. This way, you're only loading the model once which can increase inference and lower memory-usage.
 
-You can use PEFT to easily fuse/unfuse multiple adapters directly into the model weights (both UNet and text encoder) using the [`~loaders.lora_base.LoraBaseMixin.fuse_lora`] method, which can lead to a speed-up in inference and lower VRAM usage.
+You can use PEFT to easily fuse/unfuse multiple adapters directly into the model weights (both UNet and text encoder) using the [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method, which can lead to a speed-up in inference and lower VRAM usage.
 
 For example, if you have a base model and adapters loaded and set as active with the following adapter weights:
 
@@ -199,7 +199,7 @@ pipeline.load_lora_weights("lordjia/by-feng-zikai", weight_name="fengzikai_v1.0_
 pipeline.set_adapters(["ikea", "feng"], adapter_weights=[0.7, 0.8])
 ```
 
-Fuse these LoRAs into the UNet with the [`~loaders.lora_base.LoraBaseMixin.fuse_lora`] method. The `lora_scale` parameter controls how much to scale the output by with the LoRA weights. It is important to make the `lora_scale` adjustments in the [`~loaders.lora_base.LoraBaseMixin.fuse_lora`] method because it won’t work if you try to pass `scale` to the `cross_attention_kwargs` in the pipeline.
+Fuse these LoRAs into the UNet with the [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method. The `lora_scale` parameter controls how much to scale the output by with the LoRA weights. It is important to make the `lora_scale` adjustments in the [`~loaders.StableDiffusionLoraLoaderMixin.fuse_lora`] method because it won’t work if you try to pass `scale` to the `cross_attention_kwargs` in the pipeline.
 
 ```py
 pipeline.fuse_lora(adapter_names=["ikea", "feng"], lora_scale=1.0)
@@ -226,7 +226,7 @@ image = pipeline("A bowl of ramen shaped like a cute kawaii bear, by Feng Zikai"
 image
 ```
 
-You can call [`~~loaders.lora_base.LoraBaseMixin.unfuse_lora`] to restore the original model's weights (for example, if you want to use a different `lora_scale` value). However, this only works if you've only fused one LoRA adapter to the original model. If you've fused multiple LoRAs, you'll need to reload the model.
+You can call [`~loaders.StableDiffusionLoraLoaderMixin.unfuse_lora`] to restore the original model's weights (for example, if you want to use a different `lora_scale` value). However, this only works if you've only fused one LoRA adapter to the original model. If you've fused multiple LoRAs, you'll need to reload the model.
 
 ```py
 pipeline.unfuse_lora()

examples/advanced_diffusion_training/train_dreambooth_lora_sd15_advanced.py

@@ -71,7 +71,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/advanced_diffusion_training/train_dreambooth_lora_sdxl_advanced.py

@@ -79,7 +79,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/community/README.md

@@ -71,7 +71,6 @@ Please also check out our [Community Scripts](https://github.com/huggingface/dif
 | Stable Diffusion BoxDiff Pipeline | Training-free controlled generation with bounding boxes using [BoxDiff](https://github.com/showlab/BoxDiff) | [Stable Diffusion BoxDiff Pipeline](#stable-diffusion-boxdiff) | - | [Jingyang Zhang](https://github.com/zjysteven/) |
 |   FRESCO V2V Pipeline                                                                                                    | Implementation of [[CVPR 2024] FRESCO: Spatial-Temporal Correspondence for Zero-Shot Video Translation](https://arxiv.org/abs/2403.12962)                                                                                                                                                                                                                                                                                                                                                                                                                                      | [FRESCO V2V Pipeline](#fresco)      | - |              [Yifan Zhou](https://github.com/SingleZombie) |
 | AnimateDiff IPEX Pipeline | Accelerate AnimateDiff inference pipeline with BF16/FP32 precision on Intel Xeon CPUs with [IPEX](https://github.com/intel/intel-extension-for-pytorch) | [AnimateDiff on IPEX](#animatediff-on-ipex) | - | [Dan Li](https://github.com/ustcuna/) |
-| HunyuanDiT Differential Diffusion Pipeline | Applies [Differential Diffsuion](https://github.com/exx8/differential-diffusion) to [HunyuanDiT](https://github.com/huggingface/diffusers/pull/8240). | [HunyuanDiT with Differential Diffusion](#hunyuandit-with-differential-diffusion) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1v44a5fpzyr4Ffr4v2XBQ7BajzG874N4P?usp=sharing) | [Monjoy Choudhury](https://github.com/MnCSSJ4x) |
 
 To load a custom pipeline you just need to pass the `custom_pipeline` argument to `DiffusionPipeline`, as one of the files in `diffusers/examples/community`. Feel free to send a PR with your own pipelines, we will merge them quickly.
 
@@ -1647,6 +1646,7 @@ from diffusers import DiffusionPipeline
 scheduler = DDIMScheduler.from_pretrained("stabilityai/stable-diffusion-2-1",
                                             subfolder="scheduler")
 
+
 pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1",
                                             custom_pipeline="stable_diffusion_tensorrt_img2img",
                                             variant='fp16',
@@ -1661,6 +1661,7 @@ pipe = pipe.to("cuda")
 url = "https://pajoca.com/wp-content/uploads/2022/09/tekito-yamakawa-1.png"
 response = requests.get(url)
 input_image = Image.open(BytesIO(response.content)).convert("RGB")
+
 prompt = "photorealistic new zealand hills"
 image = pipe(prompt, image=input_image, strength=0.75,).images[0]
 image.save('tensorrt_img2img_new_zealand_hills.png')
@@ -4208,52 +4209,6 @@ print("Latency of AnimateDiffPipelineIpex--fp32", latency, "s for total", step,
 latency = elapsed_time(pipe4, num_inference_steps=step)
 print("Latency of AnimateDiffPipeline--fp32",latency, "s for total", step, "steps")
 ```
-### HunyuanDiT with Differential Diffusion
-
-#### Usage
-
-```python
-import torch
-from diffusers import FlowMatchEulerDiscreteScheduler
-from diffusers.utils import load_image
-from PIL import Image
-from torchvision import transforms
-
-from pipeline_hunyuandit_differential_img2img import (
-    HunyuanDiTDifferentialImg2ImgPipeline,
-)
-
-
-pipe = HunyuanDiTDifferentialImg2ImgPipeline.from_pretrained(
-    "Tencent-Hunyuan/HunyuanDiT-Diffusers", torch_dtype=torch.float16
-).to("cuda")
-
-
-source_image = load_image(
-    "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/differential/20240329211129_4024911930.png"
-)
-map = load_image(
-    "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/differential/gradient_mask_2.png"
-)
-prompt = "a green pear"
-negative_prompt = "blurry"
-
-image = pipe(
-    prompt=prompt,
-    negative_prompt=negative_prompt,
-    image=source_image,
-    num_inference_steps=28,
-    guidance_scale=4.5,
-    strength=1.0,
-    map=map,
-).images[0]
-```
-
-| ![Gradient](https://github.com/user-attachments/assets/e38ce4d5-1ae6-4df0-ab43-adc1b45716b5) | ![Input](https://github.com/user-attachments/assets/9c95679c-e9d7-4f5a-90d6-560203acd6b3) | ![Output](https://github.com/user-attachments/assets/5313ff64-a0c4-418b-8b55-a38f1a5e7532) |
-| ------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
-| Gradient                                                                                   | Input                                                                                   | Output                                                                                   |
-
-A colab notebook demonstrating all results can be found [here](https://colab.research.google.com/drive/1v44a5fpzyr4Ffr4v2XBQ7BajzG874N4P?usp=sharing). Depth Maps have also been added in the same colab.
 
 # Perturbed-Attention Guidance
 
@@ -4330,4 +4285,4 @@ grid_image.save(grid_dir + "sample.png")
 
 `pag_scale` : guidance scale of PAG (ex: 5.0)
 
-`pag_applied_layers_index` : index of the layer to apply perturbation (ex: ['m0'])
+`pag_applied_layers_index` : index of the layer to apply perturbation (ex: ['m0'])

examples/community/marigold_depth_estimation.py

@@ -43,8 +43,7 @@ from diffusers.utils import BaseOutput, check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
-
+check_min_version("0.30.0")
 
 class MarigoldDepthOutput(BaseOutput):
     """

examples/consistency_distillation/train_lcm_distill_lora_sd_wds.py

@@ -73,7 +73,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/consistency_distillation/train_lcm_distill_lora_sdxl.py

@@ -66,7 +66,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/consistency_distillation/train_lcm_distill_lora_sdxl_wds.py

@@ -79,7 +79,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/consistency_distillation/train_lcm_distill_sd_wds.py

@@ -72,7 +72,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/consistency_distillation/train_lcm_distill_sdxl_wds.py

@@ -78,7 +78,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/controlnet/train_controlnet.py

@@ -60,7 +60,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/controlnet/train_controlnet_flax.py

@@ -60,7 +60,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = logging.getLogger(__name__)
 

examples/controlnet/train_controlnet_sdxl.py

@@ -61,7 +61,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 if is_torch_npu_available():

examples/custom_diffusion/train_custom_diffusion.py

@@ -63,7 +63,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/dreambooth/README_flux.md

@@ -1,195 +0,0 @@
-# DreamBooth training example for FLUX.1 [dev]
-
-[DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text2image models like stable diffusion given just a few (3~5) images of a subject.
-
-The `train_dreambooth_flux.py` script shows how to implement the training procedure and adapt it for [FLUX.1 [dev]](https://blackforestlabs.ai/announcing-black-forest-labs/). We also provide a LoRA implementation in the `train_dreambooth_lora_flux.py` script.
-> [!NOTE]
-> **Memory consumption**
->
-> Flux can be quite expensive to run on consumer hardware devices and as a result finetuning it comes with high memory requirements -
-> a LoRA with a rank of 16 (w/ all components trained) can exceed 40GB of VRAM for training.
-> For more tips & guidance on training on a resource-constrained device please visit [`@bghira`'s guide](https://github.com/bghira/SimpleTuner/blob/main/documentation/quickstart/FLUX.md)
-
-
-> [!NOTE]
-> **Gated model**
->
-> As the model is gated, before using it with diffusers you first need to go to the [FLUX.1 [dev] Hugging Face page](https://huggingface.co/black-forest-labs/FLUX.1-dev), fill in the form and accept the gate. Once you are in, you need to log in so that your system knows you’ve accepted the gate. Use the command below to log in:
-
-```bash
-huggingface-cli login
-```
-
-This will also allow us to push the trained model parameters to the Hugging Face Hub platform.
-
-## Running locally with PyTorch
-
-### Installing the dependencies
-
-Before running the scripts, make sure to install the library's training dependencies:
-
-**Important**
-
-To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
-
-```bash
-git clone https://github.com/huggingface/diffusers
-cd diffusers
-pip install -e .
-```
-
-Then cd in the `examples/dreambooth` folder and run
-```bash
-pip install -r requirements_flux.txt
-```
-
-And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
-
-```bash
-accelerate config
-```
-
-Or for a default accelerate configuration without answering questions about your environment
-
-```bash
-accelerate config default
-```
-
-Or if your environment doesn't support an interactive shell (e.g., a notebook)
-
-```python
-from accelerate.utils import write_basic_config
-write_basic_config()
-```
-
-When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups.
-Note also that we use PEFT library as backend for LoRA training, make sure to have `peft>=0.6.0` installed in your environment.
-
-
-### Dog toy example
-
-Now let's get our dataset. For this example we will use some dog images: https://huggingface.co/datasets/diffusers/dog-example.
-
-Let's first download it locally:
-
-```python
-from huggingface_hub import snapshot_download
-
-local_dir = "./dog"
-snapshot_download(
-    "diffusers/dog-example",
-    local_dir=local_dir, repo_type="dataset",
-    ignore_patterns=".gitattributes",
-)
-```
-
-This will also allow us to push the trained LoRA parameters to the Hugging Face Hub platform.
-
-Now, we can launch training using:
-
-```bash
-export MODEL_NAME="black-forest-labs/FLUX.1-dev"
-export INSTANCE_DIR="dog"
-export OUTPUT_DIR="trained-flux"
-
-accelerate launch train_dreambooth_flux.py \
-  --pretrained_model_name_or_path=$MODEL_NAME  \
-  --instance_data_dir=$INSTANCE_DIR \
-  --output_dir=$OUTPUT_DIR \
-  --mixed_precision="bf16" \
-  --instance_prompt="a photo of sks dog" \
-  --resolution=1024 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --learning_rate=1e-4 \
-  --report_to="wandb" \
-  --lr_scheduler="constant" \
-  --lr_warmup_steps=0 \
-  --max_train_steps=500 \
-  --validation_prompt="A photo of sks dog in a bucket" \
-  --validation_epochs=25 \
-  --seed="0" \
-  --push_to_hub
-```
-
-To better track our training experiments, we're using the following flags in the command above:
-
-* `report_to="wandb` will ensure the training runs are tracked on Weights and Biases. To use it, be sure to install `wandb` with `pip install wandb`.
-* `validation_prompt` and `validation_epochs` to allow the script to do a few validation inference runs. This allows us to qualitatively check if the training is progressing as expected.
-
-> [!NOTE]
-> If you want to train using long prompts with the T5 text encoder, you can use `--max_sequence_length` to set the token limit. The default is 77, but it can be increased to as high as 512. Note that this will use more resources and may slow down the training in some cases.
-
-> [!TIP]
-> You can pass `--use_8bit_adam` to reduce the memory requirements of training. Make sure to install `bitsandbytes` if you want to do so.
-
-## LoRA + DreamBooth
-
-[LoRA](https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora) is a popular parameter-efficient fine-tuning technique that allows you to achieve full-finetuning like performance but with a fraction of learnable parameters.
-
-Note also that we use PEFT library as backend for LoRA training, make sure to have `peft>=0.6.0` installed in your environment.
-
-To perform DreamBooth with LoRA, run:
-
-```bash
-export MODEL_NAME="black-forest-labs/FLUX.1-dev"
-export INSTANCE_DIR="dog"
-export OUTPUT_DIR="trained-flux-lora"
-
-accelerate launch train_dreambooth_lora_flux.py \
-  --pretrained_model_name_or_path=$MODEL_NAME  \
-  --instance_data_dir=$INSTANCE_DIR \
-  --output_dir=$OUTPUT_DIR \
-  --mixed_precision="bf16" \
-  --instance_prompt="a photo of sks dog" \
-  --resolution=512 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --learning_rate=1e-5 \
-  --report_to="wandb" \
-  --lr_scheduler="constant" \
-  --lr_warmup_steps=0 \
-  --max_train_steps=500 \
-  --validation_prompt="A photo of sks dog in a bucket" \
-  --validation_epochs=25 \
-  --seed="0" \
-  --push_to_hub
-```
-
-### Text Encoder Training
-
-Alongside the transformer, fine-tuning of the CLIP text encoder is also supported.
-To do so, just specify `--train_text_encoder` while launching training. Please keep the following points in mind:
-
-> [!NOTE]
-> FLUX.1 has 2 text encoders (CLIP L/14 and T5-v1.1-XXL).
-By enabling `--train_text_encoder`, fine-tuning of the **CLIP encoder** is performed.
-> At the moment, T5 fine-tuning is not supported and weights remain frozen when text encoder training is enabled.
-
-To perform DreamBooth LoRA with text-encoder training, run:
-```bash
-export MODEL_NAME="black-forest-labs/FLUX.1-dev"
-export OUTPUT_DIR="trained-flux-dev-dreambooth-lora"
-
-accelerate launch train_dreambooth_lora_flux.py \
-  --pretrained_model_name_or_path=$MODEL_NAME  \
-  --instance_data_dir=$INSTANCE_DIR \
-  --output_dir=$OUTPUT_DIR \
-  --mixed_precision="bf16" \
-  --train_text_encoder\
-  --instance_prompt="a photo of sks dog" \
-  --resolution=512 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --learning_rate=1e-5 \
-  --report_to="wandb" \
-  --lr_scheduler="constant" \
-  --lr_warmup_steps=0 \
-  --max_train_steps=500 \
-  --validation_prompt="A photo of sks dog in a bucket" \
-  --seed="0" \
-  --push_to_hub
-```
-
-## Other notes
-Thanks to `bghira` for their help with reviewing & insight sharing ♥️

examples/dreambooth/requirements_flux.txt

@@ -1,8 +0,0 @@
-accelerate>=0.31.0
-torchvision
-transformers>=4.41.2
-ftfy
-tensorboard
-Jinja2
-peft>=0.11.1
-sentencepiece

examples/dreambooth/test_dreambooth_flux.py

@@ -1,203 +0,0 @@
-# coding=utf-8
-# Copyright 2024 HuggingFace Inc.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import logging
-import os
-import shutil
-import sys
-import tempfile
-
-from diffusers import DiffusionPipeline, FluxTransformer2DModel
-
-
-sys.path.append("..")
-from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
-
-
-logging.basicConfig(level=logging.DEBUG)
-
-logger = logging.getLogger()
-stream_handler = logging.StreamHandler(sys.stdout)
-logger.addHandler(stream_handler)
-
-
-class DreamBoothFlux(ExamplesTestsAccelerate):
-    instance_data_dir = "docs/source/en/imgs"
-    instance_prompt = "photo"
-    pretrained_model_name_or_path = "hf-internal-testing/tiny-flux-pipe"
-    script_path = "examples/dreambooth/train_dreambooth_flux.py"
-
-    def test_dreambooth(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-                {self.script_path}
-                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
-                --instance_data_dir {self.instance_data_dir}
-                --instance_prompt {self.instance_prompt}
-                --resolution 64
-                --train_batch_size 1
-                --gradient_accumulation_steps 1
-                --max_train_steps 2
-                --learning_rate 5.0e-04
-                --scale_lr
-                --lr_scheduler constant
-                --lr_warmup_steps 0
-                --output_dir {tmpdir}
-                """.split()
-
-            run_command(self._launch_args + test_args)
-            # save_pretrained smoke test
-            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "transformer", "diffusion_pytorch_model.safetensors")))
-            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "scheduler", "scheduler_config.json")))
-
-    def test_dreambooth_checkpointing(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            # Run training script with checkpointing
-            # max_train_steps == 4, checkpointing_steps == 2
-            # Should create checkpoints at steps 2, 4
-
-            initial_run_args = f"""
-                {self.script_path}
-                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
-                --instance_data_dir {self.instance_data_dir}
-                --instance_prompt {self.instance_prompt}
-                --resolution 64
-                --train_batch_size 1
-                --gradient_accumulation_steps 1
-                --max_train_steps 4
-                --learning_rate 5.0e-04
-                --scale_lr
-                --lr_scheduler constant
-                --lr_warmup_steps 0
-                --output_dir {tmpdir}
-                --checkpointing_steps=2
-                --seed=0
-                """.split()
-
-            run_command(self._launch_args + initial_run_args)
-
-            # check can run the original fully trained output pipeline
-            pipe = DiffusionPipeline.from_pretrained(tmpdir)
-            pipe(self.instance_prompt, num_inference_steps=1)
-
-            # check checkpoint directories exist
-            self.assertTrue(os.path.isdir(os.path.join(tmpdir, "checkpoint-2")))
-            self.assertTrue(os.path.isdir(os.path.join(tmpdir, "checkpoint-4")))
-
-            # check can run an intermediate checkpoint
-            transformer = FluxTransformer2DModel.from_pretrained(tmpdir, subfolder="checkpoint-2/transformer")
-            pipe = DiffusionPipeline.from_pretrained(self.pretrained_model_name_or_path, transformer=transformer)
-            pipe(self.instance_prompt, num_inference_steps=1)
-
-            # Remove checkpoint 2 so that we can check only later checkpoints exist after resuming
-            shutil.rmtree(os.path.join(tmpdir, "checkpoint-2"))
-
-            # Run training script for 7 total steps resuming from checkpoint 4
-
-            resume_run_args = f"""
-                {self.script_path}
-                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
-                --instance_data_dir {self.instance_data_dir}
-                --instance_prompt {self.instance_prompt}
-                --resolution 64
-                --train_batch_size 1
-                --gradient_accumulation_steps 1
-                --max_train_steps 6
-                --learning_rate 5.0e-04
-                --scale_lr
-                --lr_scheduler constant
-                --lr_warmup_steps 0
-                --output_dir {tmpdir}
-                --checkpointing_steps=2
-                --resume_from_checkpoint=checkpoint-4
-                --seed=0
-                """.split()
-
-            run_command(self._launch_args + resume_run_args)
-
-            # check can run new fully trained pipeline
-            pipe = DiffusionPipeline.from_pretrained(tmpdir)
-            pipe(self.instance_prompt, num_inference_steps=1)
-
-            # check old checkpoints do not exist
-            self.assertFalse(os.path.isdir(os.path.join(tmpdir, "checkpoint-2")))
-
-            # check new checkpoints exist
-            self.assertTrue(os.path.isdir(os.path.join(tmpdir, "checkpoint-4")))
-            self.assertTrue(os.path.isdir(os.path.join(tmpdir, "checkpoint-6")))
-
-    def test_dreambooth_checkpointing_checkpoints_total_limit(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=6
-            --checkpoints_total_limit=2
-            --checkpointing_steps=2
-            """.split()
-
-            run_command(self._launch_args + test_args)
-
-            self.assertEqual(
-                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
-                {"checkpoint-4", "checkpoint-6"},
-            )
-
-    def test_dreambooth_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=4
-            --checkpointing_steps=2
-            """.split()
-
-            run_command(self._launch_args + test_args)
-
-            self.assertEqual(
-                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
-                {"checkpoint-2", "checkpoint-4"},
-            )
-
-            resume_run_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=8
-            --checkpointing_steps=2
-            --resume_from_checkpoint=checkpoint-4
-            --checkpoints_total_limit=2
-            """.split()
-
-            run_command(self._launch_args + resume_run_args)
-
-            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-6", "checkpoint-8"})

examples/dreambooth/test_dreambooth_lora_flux.py

@@ -1,165 +0,0 @@
-# coding=utf-8
-# Copyright 2024 HuggingFace Inc.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import logging
-import os
-import sys
-import tempfile
-
-import safetensors
-
-
-sys.path.append("..")
-from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
-
-
-logging.basicConfig(level=logging.DEBUG)
-
-logger = logging.getLogger()
-stream_handler = logging.StreamHandler(sys.stdout)
-logger.addHandler(stream_handler)
-
-
-class DreamBoothLoRAFlux(ExamplesTestsAccelerate):
-    instance_data_dir = "docs/source/en/imgs"
-    instance_prompt = "photo"
-    pretrained_model_name_or_path = "hf-internal-testing/tiny-flux-pipe"
-    script_path = "examples/dreambooth/train_dreambooth_lora_flux.py"
-
-    def test_dreambooth_lora_flux(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-                {self.script_path}
-                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
-                --instance_data_dir {self.instance_data_dir}
-                --instance_prompt {self.instance_prompt}
-                --resolution 64
-                --train_batch_size 1
-                --gradient_accumulation_steps 1
-                --max_train_steps 2
-                --learning_rate 5.0e-04
-                --scale_lr
-                --lr_scheduler constant
-                --lr_warmup_steps 0
-                --output_dir {tmpdir}
-                """.split()
-
-            run_command(self._launch_args + test_args)
-            # save_pretrained smoke test
-            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
-
-            # make sure the state_dict has the correct naming in the parameters.
-            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
-            is_lora = all("lora" in k for k in lora_state_dict.keys())
-            self.assertTrue(is_lora)
-
-            # when not training the text encoder, all the parameters in the state dict should start
-            # with `"transformer"` in their names.
-            starts_with_transformer = all(key.startswith("transformer") for key in lora_state_dict.keys())
-            self.assertTrue(starts_with_transformer)
-
-    def test_dreambooth_lora_text_encoder_flux(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-                {self.script_path}
-                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
-                --instance_data_dir {self.instance_data_dir}
-                --instance_prompt {self.instance_prompt}
-                --resolution 64
-                --train_batch_size 1
-                --train_text_encoder
-                --gradient_accumulation_steps 1
-                --max_train_steps 2
-                --learning_rate 5.0e-04
-                --scale_lr
-                --lr_scheduler constant
-                --lr_warmup_steps 0
-                --output_dir {tmpdir}
-                """.split()
-
-            run_command(self._launch_args + test_args)
-            # save_pretrained smoke test
-            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
-
-            # make sure the state_dict has the correct naming in the parameters.
-            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
-            is_lora = all("lora" in k for k in lora_state_dict.keys())
-            self.assertTrue(is_lora)
-
-            starts_with_expected_prefix = all(
-                (key.startswith("transformer") or key.startswith("text_encoder")) for key in lora_state_dict.keys()
-            )
-            self.assertTrue(starts_with_expected_prefix)
-
-    def test_dreambooth_lora_flux_checkpointing_checkpoints_total_limit(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=6
-            --checkpoints_total_limit=2
-            --checkpointing_steps=2
-            """.split()
-
-            run_command(self._launch_args + test_args)
-
-            self.assertEqual(
-                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
-                {"checkpoint-4", "checkpoint-6"},
-            )
-
-    def test_dreambooth_lora_flux_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            test_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=4
-            --checkpointing_steps=2
-            """.split()
-
-            run_command(self._launch_args + test_args)
-
-            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-2", "checkpoint-4"})
-
-            resume_run_args = f"""
-            {self.script_path}
-            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
-            --instance_data_dir={self.instance_data_dir}
-            --output_dir={tmpdir}
-            --instance_prompt={self.instance_prompt}
-            --resolution=64
-            --train_batch_size=1
-            --gradient_accumulation_steps=1
-            --max_train_steps=8
-            --checkpointing_steps=2
-            --resume_from_checkpoint=checkpoint-4
-            --checkpoints_total_limit=2
-            """.split()
-
-            run_command(self._launch_args + resume_run_args)
-
-            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-6", "checkpoint-8"})

examples/dreambooth/train_dreambooth.py

@@ -63,7 +63,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/dreambooth/train_dreambooth_flax.py

@@ -35,7 +35,7 @@ from diffusers.utils import check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 # Cache compiled models across invocations of this script.
 cc.initialize_cache(os.path.expanduser("~/.cache/jax/compilation_cache"))

examples/dreambooth/train_dreambooth_lora.py

@@ -70,7 +70,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/dreambooth/train_dreambooth_lora_sd3.py

@@ -72,7 +72,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 
@@ -1454,7 +1454,7 @@ def main(args):
             )
 
     # Clear the memory here
-    if not args.train_text_encoder and not train_dataset.custom_instance_prompts:
+    if not args.train_text_encoder and train_dataset.custom_instance_prompts:
         del tokenizers, text_encoders
         # Explicitly delete the objects as well, otherwise only the lists are deleted and the original references remain, preventing garbage collection
         del text_encoder_one, text_encoder_two, text_encoder_three

examples/dreambooth/train_dreambooth_lora_sdxl.py

@@ -78,7 +78,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/dreambooth/train_dreambooth_sd3.py

@@ -64,7 +64,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/instruct_pix2pix/train_instruct_pix2pix.py

@@ -57,7 +57,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py

@@ -60,7 +60,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/kandinsky2_2/text_to_image/train_text_to_image_decoder.py

@@ -52,7 +52,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/kandinsky2_2/text_to_image/train_text_to_image_lora_decoder.py

@@ -46,7 +46,7 @@ from diffusers.utils import check_min_version, is_wandb_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/kandinsky2_2/text_to_image/train_text_to_image_lora_prior.py

@@ -46,7 +46,7 @@ from diffusers.utils import check_min_version, is_wandb_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py

@@ -51,7 +51,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/research_projects/realfill/requirements.txt

@@ -2,8 +2,8 @@ diffusers==0.20.1
 accelerate==0.23.0
 transformers==4.38.0
 peft==0.5.0
-torch==2.2.0
+torch==2.0.1
 torchvision>=0.16
 ftfy==6.1.1
 tensorboard==2.14.0
-Jinja2==3.1.4
+Jinja2==3.1.3

examples/t2i_adapter/train_t2i_adapter_sdxl.py

@@ -60,7 +60,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/text_to_image/train_text_to_image.py

@@ -57,7 +57,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -826,22 +826,17 @@ def main():
     )
 
     # Scheduler and math around the number of training steps.
-    # Check the PR https://github.com/huggingface/diffusers/pull/8312 for detailed explanation.
-    num_warmup_steps_for_scheduler = args.lr_warmup_steps * accelerator.num_processes
+    overrode_max_train_steps = False
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
     if args.max_train_steps is None:
-        len_train_dataloader_after_sharding = math.ceil(len(train_dataloader) / accelerator.num_processes)
-        num_update_steps_per_epoch = math.ceil(len_train_dataloader_after_sharding / args.gradient_accumulation_steps)
-        num_training_steps_for_scheduler = (
-            args.num_train_epochs * num_update_steps_per_epoch * accelerator.num_processes
-        )
-    else:
-        num_training_steps_for_scheduler = args.max_train_steps * accelerator.num_processes
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        overrode_max_train_steps = True
 
     lr_scheduler = get_scheduler(
         args.lr_scheduler,
         optimizer=optimizer,
-        num_warmup_steps=num_warmup_steps_for_scheduler,
-        num_training_steps=num_training_steps_for_scheduler,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
     )
 
     # Prepare everything with our `accelerator`.
@@ -871,14 +866,8 @@ def main():
 
     # We need to recalculate our total training steps as the size of the training dataloader may have changed.
     num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if args.max_train_steps is None:
+    if overrode_max_train_steps:
         args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        if num_training_steps_for_scheduler != args.max_train_steps * accelerator.num_processes:
-            logger.warning(
-                f"The length of the 'train_dataloader' after 'accelerator.prepare' ({len(train_dataloader)}) does not match "
-                f"the expected length ({len_train_dataloader_after_sharding}) when the learning rate scheduler was created. "
-                f"This inconsistency may result in the learning rate scheduler not functioning properly."
-            )
     # Afterwards we recalculate our number of training epochs
     args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
 

examples/text_to_image/train_text_to_image_flax.py

@@ -49,7 +49,7 @@ from diffusers.utils import check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = logging.getLogger(__name__)
 

examples/text_to_image/train_text_to_image_lora.py

@@ -56,7 +56,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/text_to_image/train_text_to_image_lora_sdxl.py

@@ -68,7 +68,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 if is_torch_npu_available():
@@ -478,7 +478,7 @@ def parse_args(input_args=None):
     parser.add_argument(
         "--debug_loss",
         action="store_true",
-        help="debug loss for each image, if filenames are available in the dataset",
+        help="debug loss for each image, if filenames are awailable in the dataset",
     )
 
     if input_args is not None:

examples/text_to_image/train_text_to_image_sdxl.py

@@ -55,7 +55,7 @@ from diffusers.utils.torch_utils import is_compiled_module
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 if is_torch_npu_available():

examples/textual_inversion/README.md

@@ -109,9 +109,6 @@ import torch
 model_id = "path-to-your-trained-model"
 pipe = StableDiffusionPipeline.from_pretrained(model_id,torch_dtype=torch.float16).to("cuda")
 
-repo_id_embeds = "path-to-your-learned-embeds"
-pipe.load_textual_inversion(repo_id_embeds)
-
 prompt = "A <cat-toy> backpack"
 
 image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]

examples/textual_inversion/README_sdxl.md

@@ -23,25 +23,4 @@ accelerate launch textual_inversion_sdxl.py \
   --output_dir="./textual_inversion_cat_sdxl"
 ```
 
-Training of both text encoders is supported.
-
-### Inference Example
-
-Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionXLPipeline`.
-Make sure to include the `placeholder_token` in your prompt.
-
-```python
-from diffusers import StableDiffusionXLPipeline
-import torch
-
-model_id = "./textual_inversion_cat_sdxl"
-pipe = StableDiffusionXLPipeline.from_pretrained(model_id,torch_dtype=torch.float16).to("cuda")
-
-prompt = "A <cat-toy> backpack"
-
-image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
-image.save("cat-backpack.png")
-
-image = pipe(prompt="", prompt_2=prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
-image.save("cat-backpack-prompt_2.png")
-```
+For now, only training of the first text encoder is supported.

examples/textual_inversion/textual_inversion.py

@@ -81,7 +81,7 @@ else:
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 

examples/textual_inversion/textual_inversion_flax.py

@@ -56,7 +56,7 @@ else:
 # ------------------------------------------------------------------------------
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = logging.getLogger(__name__)
 

examples/textual_inversion/textual_inversion_sdxl.py

@@ -76,7 +76,7 @@ else:
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__)
 
@@ -135,7 +135,7 @@ def log_validation(
     pipeline = DiffusionPipeline.from_pretrained(
         args.pretrained_model_name_or_path,
         text_encoder=accelerator.unwrap_model(text_encoder_1),
-        text_encoder_2=accelerator.unwrap_model(text_encoder_2),
+        text_encoder_2=text_encoder_2,
         tokenizer=tokenizer_1,
         tokenizer_2=tokenizer_2,
         unet=unet,
@@ -678,54 +678,36 @@ def main():
             f"The tokenizer already contains the token {args.placeholder_token}. Please pass a different"
             " `placeholder_token` that is not already in the tokenizer."
         )
-    num_added_tokens = tokenizer_2.add_tokens(placeholder_tokens)
-    if num_added_tokens != args.num_vectors:
-        raise ValueError(
-            f"The 2nd tokenizer already contains the token {args.placeholder_token}. Please pass a different"
-            " `placeholder_token` that is not already in the tokenizer."
-        )
 
     # Convert the initializer_token, placeholder_token to ids
     token_ids = tokenizer_1.encode(args.initializer_token, add_special_tokens=False)
-    token_ids_2 = tokenizer_2.encode(args.initializer_token, add_special_tokens=False)
-
     # Check if initializer_token is a single token or a sequence of tokens
-    if len(token_ids) > 1 or len(token_ids_2) > 1:
+    if len(token_ids) > 1:
         raise ValueError("The initializer token must be a single token.")
 
     initializer_token_id = token_ids[0]
     placeholder_token_ids = tokenizer_1.convert_tokens_to_ids(placeholder_tokens)
-    initializer_token_id_2 = token_ids_2[0]
-    placeholder_token_ids_2 = tokenizer_2.convert_tokens_to_ids(placeholder_tokens)
 
     # Resize the token embeddings as we are adding new special tokens to the tokenizer
     text_encoder_1.resize_token_embeddings(len(tokenizer_1))
-    text_encoder_2.resize_token_embeddings(len(tokenizer_2))
 
     # Initialise the newly added placeholder token with the embeddings of the initializer token
     token_embeds = text_encoder_1.get_input_embeddings().weight.data
-    token_embeds_2 = text_encoder_2.get_input_embeddings().weight.data
     with torch.no_grad():
         for token_id in placeholder_token_ids:
             token_embeds[token_id] = token_embeds[initializer_token_id].clone()
-        for token_id in placeholder_token_ids_2:
-            token_embeds_2[token_id] = token_embeds_2[initializer_token_id_2].clone()
 
     # Freeze vae and unet
     vae.requires_grad_(False)
     unet.requires_grad_(False)
-
+    text_encoder_2.requires_grad_(False)
     # Freeze all parameters except for the token embeddings in text encoder
     text_encoder_1.text_model.encoder.requires_grad_(False)
     text_encoder_1.text_model.final_layer_norm.requires_grad_(False)
     text_encoder_1.text_model.embeddings.position_embedding.requires_grad_(False)
-    text_encoder_2.text_model.encoder.requires_grad_(False)
-    text_encoder_2.text_model.final_layer_norm.requires_grad_(False)
-    text_encoder_2.text_model.embeddings.position_embedding.requires_grad_(False)
 
     if args.gradient_checkpointing:
         text_encoder_1.gradient_checkpointing_enable()
-        text_encoder_2.gradient_checkpointing_enable()
 
     if args.enable_xformers_memory_efficient_attention:
         if is_xformers_available():
@@ -764,11 +746,7 @@ def main():
         optimizer_class = torch.optim.AdamW
 
     optimizer = optimizer_class(
-        # only optimize the embeddings
-        [
-            text_encoder_1.text_model.embeddings.token_embedding.weight,
-            text_encoder_2.text_model.embeddings.token_embedding.weight,
-        ],
+        text_encoder_1.get_input_embeddings().parameters(),  # only optimize the embeddings
         lr=args.learning_rate,
         betas=(args.adam_beta1, args.adam_beta2),
         weight_decay=args.adam_weight_decay,
@@ -808,10 +786,9 @@ def main():
     )
 
     text_encoder_1.train()
-    text_encoder_2.train()
     # Prepare everything with our `accelerator`.
-    text_encoder_1, text_encoder_2, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        text_encoder_1, text_encoder_2, optimizer, train_dataloader, lr_scheduler
+    text_encoder_1, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder_1, optimizer, train_dataloader, lr_scheduler
     )
 
     # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
@@ -889,13 +866,11 @@ def main():
 
     # keep original embeddings as reference
     orig_embeds_params = accelerator.unwrap_model(text_encoder_1).get_input_embeddings().weight.data.clone()
-    orig_embeds_params_2 = accelerator.unwrap_model(text_encoder_2).get_input_embeddings().weight.data.clone()
 
     for epoch in range(first_epoch, args.num_train_epochs):
         text_encoder_1.train()
-        text_encoder_2.train()
         for step, batch in enumerate(train_dataloader):
-            with accelerator.accumulate([text_encoder_1, text_encoder_2]):
+            with accelerator.accumulate(text_encoder_1):
                 # Convert images to latent space
                 latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample().detach()
                 latents = latents * vae.config.scaling_factor
@@ -917,7 +892,9 @@ def main():
                     .hidden_states[-2]
                     .to(dtype=weight_dtype)
                 )
-                encoder_output_2 = text_encoder_2(batch["input_ids_2"], output_hidden_states=True)
+                encoder_output_2 = text_encoder_2(
+                    batch["input_ids_2"].reshape(batch["input_ids_1"].shape[0], -1), output_hidden_states=True
+                )
                 encoder_hidden_states_2 = encoder_output_2.hidden_states[-2].to(dtype=weight_dtype)
                 original_size = [
                     (batch["original_size"][0][i].item(), batch["original_size"][1][i].item())
@@ -961,16 +938,11 @@ def main():
                 # Let's make sure we don't update any embedding weights besides the newly added token
                 index_no_updates = torch.ones((len(tokenizer_1),), dtype=torch.bool)
                 index_no_updates[min(placeholder_token_ids) : max(placeholder_token_ids) + 1] = False
-                index_no_updates_2 = torch.ones((len(tokenizer_2),), dtype=torch.bool)
-                index_no_updates_2[min(placeholder_token_ids_2) : max(placeholder_token_ids_2) + 1] = False
 
                 with torch.no_grad():
                     accelerator.unwrap_model(text_encoder_1).get_input_embeddings().weight[
                         index_no_updates
                     ] = orig_embeds_params[index_no_updates]
-                    accelerator.unwrap_model(text_encoder_2).get_input_embeddings().weight[
-                        index_no_updates_2
-                    ] = orig_embeds_params_2[index_no_updates_2]
 
             # Checks if the accelerator has performed an optimization step behind the scenes
             if accelerator.sync_gradients:
@@ -988,16 +960,6 @@ def main():
                         save_path,
                         safe_serialization=True,
                     )
-                    weight_name = f"learned_embeds_2-steps-{global_step}.safetensors"
-                    save_path = os.path.join(args.output_dir, weight_name)
-                    save_progress(
-                        text_encoder_2,
-                        placeholder_token_ids_2,
-                        accelerator,
-                        args,
-                        save_path,
-                        safe_serialization=True,
-                    )
 
                 if accelerator.is_main_process:
                     if global_step % args.checkpointing_steps == 0:
@@ -1072,7 +1034,7 @@ def main():
             pipeline = DiffusionPipeline.from_pretrained(
                 args.pretrained_model_name_or_path,
                 text_encoder=accelerator.unwrap_model(text_encoder_1),
-                text_encoder_2=accelerator.unwrap_model(text_encoder_2),
+                text_encoder_2=text_encoder_2,
                 vae=vae,
                 unet=unet,
                 tokenizer=tokenizer_1,
@@ -1090,16 +1052,6 @@ def main():
             save_path,
             safe_serialization=True,
         )
-        weight_name = "learned_embeds_2.safetensors"
-        save_path = os.path.join(args.output_dir, weight_name)
-        save_progress(
-            text_encoder_2,
-            placeholder_token_ids_2,
-            accelerator,
-            args,
-            save_path,
-            safe_serialization=True,
-        )
 
         if args.push_to_hub:
             save_model_card(

examples/unconditional_image_generation/train_unconditional.py

@@ -29,7 +29,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/vqgan/train_vqgan.py

@@ -50,7 +50,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/wuerstchen/text_to_image/train_text_to_image_lora_prior.py

@@ -50,7 +50,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/wuerstchen/text_to_image/train_text_to_image_prior.py

@@ -51,7 +51,7 @@ if is_wandb_available():
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.31.0.dev0")
+check_min_version("0.30.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

scripts/convert_cogvideox_to_diffusers.py

@@ -86,6 +86,9 @@ TRANSFORMER_SPECIAL_KEYS_REMAP = {
     "key_layernorm_list": reassign_query_key_layernorm_inplace,
     "adaln_layer.adaLN_modulations": reassign_adaln_norm_inplace,
     "embed_tokens": remove_keys_inplace,
+    "freqs_sin": remove_keys_inplace,
+    "freqs_cos": remove_keys_inplace,
+    "position_embedding": remove_keys_inplace,
 }
 
 VAE_KEYS_RENAME_DICT = {
@@ -123,11 +126,21 @@ def update_state_dict_inplace(state_dict: Dict[str, Any], old_key: str, new_key:
     state_dict[new_key] = state_dict.pop(old_key)
 
 
-def convert_transformer(ckpt_path: str):
+def convert_transformer(
+    ckpt_path: str,
+    num_layers: int,
+    num_attention_heads: int,
+    use_rotary_positional_embeddings: bool,
+    dtype: torch.dtype,
+):
     PREFIX_KEY = "model.diffusion_model."
 
     original_state_dict = get_state_dict(torch.load(ckpt_path, map_location="cpu", mmap=True))
-    transformer = CogVideoXTransformer3DModel()
+    transformer = CogVideoXTransformer3DModel(
+        num_layers=num_layers,
+        num_attention_heads=num_attention_heads,
+        use_rotary_positional_embeddings=use_rotary_positional_embeddings,
+    ).to(dtype=dtype)
 
     for key in list(original_state_dict.keys()):
         new_key = key[len(PREFIX_KEY) :]
@@ -145,9 +158,9 @@ def convert_transformer(ckpt_path: str):
     return transformer
 
 
-def convert_vae(ckpt_path: str):
+def convert_vae(ckpt_path: str, scaling_factor: float, dtype: torch.dtype):
     original_state_dict = get_state_dict(torch.load(ckpt_path, map_location="cpu", mmap=True))
-    vae = AutoencoderKLCogVideoX()
+    vae = AutoencoderKLCogVideoX(scaling_factor=scaling_factor).to(dtype=dtype)
 
     for key in list(original_state_dict.keys()):
         new_key = key[:]
@@ -172,13 +185,26 @@ def get_args():
     )
     parser.add_argument("--vae_ckpt_path", type=str, default=None, help="Path to original vae checkpoint")
     parser.add_argument("--output_path", type=str, required=True, help="Path where converted model should be saved")
-    parser.add_argument("--fp16", action="store_true", default=True, help="Whether to save the model weights in fp16")
+    parser.add_argument("--fp16", action="store_true", default=False, help="Whether to save the model weights in fp16")
+    parser.add_argument("--bf16", action="store_true", default=False, help="Whether to save the model weights in bf16")
     parser.add_argument(
         "--push_to_hub", action="store_true", default=False, help="Whether to push to HF Hub after saving"
     )
     parser.add_argument(
         "--text_encoder_cache_dir", type=str, default=None, help="Path to text encoder cache directory"
     )
+    # For CogVideoX-2B, num_layers is 30. For 5B, it is 42
+    parser.add_argument("--num_layers", type=int, default=30, help="Number of transformer blocks")
+    # For CogVideoX-2B, num_attention_heads is 30. For 5B, it is 48
+    parser.add_argument("--num_attention_heads", type=int, default=30, help="Number of attention heads")
+    # For CogVideoX-2B, use_rotary_positional_embeddings is False. For 5B, it is True
+    parser.add_argument(
+        "--use_rotary_positional_embeddings", action="store_true", default=False, help="Whether to use RoPE or not"
+    )
+    # For CogVideoX-2B, scaling_factor is 1.15258426. For 5B, it is 0.7
+    parser.add_argument("--scaling_factor", type=float, default=1.15258426, help="Scaling factor in the VAE")
+    # For CogVideoX-2B, snr_shift_scale is 3.0. For 5B, it is 1.0
+    parser.add_argument("--snr_shift_scale", type=float, default=3.0, help="Scaling factor in the VAE")
     return parser.parse_args()
 
 
@@ -188,18 +214,33 @@ if __name__ == "__main__":
     transformer = None
     vae = None
 
+    if args.fp16 and args.bf16:
+        raise ValueError("You cannot pass both --fp16 and --bf16 at the same time.")
+
+    dtype = torch.float16 if args.fp16 else torch.bfloat16 if args.bf16 else torch.float32
+
     if args.transformer_ckpt_path is not None:
-        transformer = convert_transformer(args.transformer_ckpt_path)
+        transformer = convert_transformer(
+            args.transformer_ckpt_path,
+            args.num_layers,
+            args.num_attention_heads,
+            args.use_rotary_positional_embeddings,
+            dtype,
+        )
     if args.vae_ckpt_path is not None:
-        vae = convert_vae(args.vae_ckpt_path)
+        vae = convert_vae(args.vae_ckpt_path, args.scaling_factor, dtype)
 
     text_encoder_id = "google/t5-v1_1-xxl"
     tokenizer = T5Tokenizer.from_pretrained(text_encoder_id, model_max_length=TOKENIZER_MAX_LENGTH)
     text_encoder = T5EncoderModel.from_pretrained(text_encoder_id, cache_dir=args.text_encoder_cache_dir)
 
+    # Apparently, the conversion does not work any more without this :shrug:
+    for param in text_encoder.parameters():
+        param.data = param.data.contiguous()
+
     scheduler = CogVideoXDDIMScheduler.from_config(
         {
-            "snr_shift_scale": 3.0,
+            "snr_shift_scale": args.snr_shift_scale,
             "beta_end": 0.012,
             "beta_schedule": "scaled_linear",
             "beta_start": 0.00085,
@@ -208,7 +249,7 @@ if __name__ == "__main__":
             "prediction_type": "v_prediction",
             "rescale_betas_zero_snr": True,
             "set_alpha_to_one": True,
-            "timestep_spacing": "linspace",
+            "timestep_spacing": "trailing",
         }
     )
 
@@ -218,5 +259,10 @@ if __name__ == "__main__":
 
     if args.fp16:
         pipe = pipe.to(dtype=torch.float16)
+    if args.bf16:
+        pipe = pipe.to(dtype=torch.bfloat16)
 
+    # We don't use variant here because the model must be run in fp16 (2B) or bf16 (5B). It would be weird
+    # for users to specify variant when the default is not fp32 and they want to run with the correct default (which
+    # is either fp16/bf16 here).
     pipe.save_pretrained(args.output_path, safe_serialization=True, push_to_hub=args.push_to_hub)

setup.py

@@ -254,7 +254,7 @@ version_range_max = max(sys.version_info[1], 10) + 1
 
 setup(
     name="diffusers",
-    version="0.31.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="0.30.2",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     description="State-of-the-art diffusion in PyTorch and JAX.",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/diffusers/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.31.0.dev0"
+__version__ = "0.30.2"
 
 from typing import TYPE_CHECKING
 
@@ -88,7 +88,6 @@ else:
             "ControlNetModel",
             "ControlNetXSAdapter",
             "DiTTransformer2DModel",
-            "FluxControlNetModel",
             "FluxTransformer2DModel",
             "HunyuanDiT2DControlNetModel",
             "HunyuanDiT2DModel",
@@ -255,7 +254,6 @@ else:
             "CLIPImageProjection",
             "CogVideoXPipeline",
             "CycleDiffusionPipeline",
-            "FluxControlNetPipeline",
             "FluxPipeline",
             "HunyuanDiTControlNetPipeline",
             "HunyuanDiTPAGPipeline",
@@ -310,7 +308,6 @@ else:
             "StableCascadeCombinedPipeline",
             "StableCascadeDecoderPipeline",
             "StableCascadePriorPipeline",
-            "StableDiffusion3ControlNetInpaintingPipeline",
             "StableDiffusion3ControlNetPipeline",
             "StableDiffusion3Img2ImgPipeline",
             "StableDiffusion3InpaintPipeline",
@@ -552,7 +549,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             ControlNetModel,
             ControlNetXSAdapter,
             DiTTransformer2DModel,
-            FluxControlNetModel,
             FluxTransformer2DModel,
             HunyuanDiT2DControlNetModel,
             HunyuanDiT2DModel,
@@ -697,7 +693,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             CLIPImageProjection,
             CogVideoXPipeline,
             CycleDiffusionPipeline,
-            FluxControlNetPipeline,
             FluxPipeline,
             HunyuanDiTControlNetPipeline,
             HunyuanDiTPAGPipeline,

src/diffusers/loaders/ip_adapter.py

@@ -208,6 +208,8 @@ class IPAdapterMixin:
                             pretrained_model_name_or_path_or_dict,
                             subfolder=image_encoder_subfolder,
                             low_cpu_mem_usage=low_cpu_mem_usage,
+                            cache_dir=cache_dir,
+                            local_files_only=local_files_only,
                         ).to(self.device, dtype=self.dtype)
                         self.register_modules(image_encoder=image_encoder)
                     else:
@@ -222,11 +224,7 @@ class IPAdapterMixin:
 
             # create feature extractor if it has not been registered to the pipeline yet
             if hasattr(self, "feature_extractor") and getattr(self, "feature_extractor", None) is None:
-                # FaceID IP adapters don't need the image encoder so it's not present, in this case we default to 224
-                default_clip_size = 224
-                clip_image_size = (
-                    self.image_encoder.config.image_size if self.image_encoder is not None else default_clip_size
-                )
+                clip_image_size = self.image_encoder.config.image_size
                 feature_extractor = CLIPImageProcessor(size=clip_image_size, crop_size=clip_image_size)
                 self.register_modules(feature_extractor=feature_extractor)
 

src/diffusers/loaders/lora_pipeline.py

@@ -1489,10 +1489,10 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
 
     @classmethod
     @validate_hf_hub_args
-    # Copied from diffusers.loaders.lora_pipeline.SD3LoraLoaderMixin.lora_state_dict
     def lora_state_dict(
         cls,
         pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]],
+        return_alphas: bool = False,
         **kwargs,
     ):
         r"""
@@ -1577,7 +1577,26 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
             allow_pickle=allow_pickle,
         )
 
-        return state_dict
+        # For state dicts like
+        # https://huggingface.co/TheLastBen/Jon_Snow_Flux_LoRA
+        keys = list(state_dict.keys())
+        network_alphas = {}
+        for k in keys:
+            if "alpha" in k:
+                alpha_value = state_dict.get(k)
+                if (torch.is_tensor(alpha_value) and torch.is_floating_point(alpha_value)) or isinstance(
+                    alpha_value, float
+                ):
+                    network_alphas[k] = state_dict.pop(k)
+                else:
+                    raise ValueError(
+                        f"The alpha key ({k}) seems to be incorrect. If you think this error is unexpected, please open as issue."
+                    )
+
+        if return_alphas:
+            return state_dict, network_alphas
+        else:
+            return state_dict
 
     def load_lora_weights(
         self, pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]], adapter_name=None, **kwargs
@@ -1611,7 +1630,9 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
             pretrained_model_name_or_path_or_dict = pretrained_model_name_or_path_or_dict.copy()
 
         # First, ensure that the checkpoint is a compatible one and can be successfully loaded.
-        state_dict = self.lora_state_dict(pretrained_model_name_or_path_or_dict, **kwargs)
+        state_dict, network_alphas = self.lora_state_dict(
+            pretrained_model_name_or_path_or_dict, return_alphas=True, **kwargs
+        )
 
         is_correct_format = all("lora" in key or "dora_scale" in key for key in state_dict.keys())
         if not is_correct_format:
@@ -1619,6 +1640,7 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
 
         self.load_lora_into_transformer(
             state_dict,
+            network_alphas=network_alphas,
             transformer=getattr(self, self.transformer_name) if not hasattr(self, "transformer") else self.transformer,
             adapter_name=adapter_name,
             _pipeline=self,
@@ -1628,7 +1650,7 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
         if len(text_encoder_state_dict) > 0:
             self.load_lora_into_text_encoder(
                 text_encoder_state_dict,
-                network_alphas=None,
+                network_alphas=network_alphas,
                 text_encoder=self.text_encoder,
                 prefix="text_encoder",
                 lora_scale=self.lora_scale,
@@ -1637,8 +1659,7 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
             )
 
     @classmethod
-    # Copied from diffusers.loaders.lora_pipeline.SD3LoraLoaderMixin.load_lora_into_transformer
-    def load_lora_into_transformer(cls, state_dict, transformer, adapter_name=None, _pipeline=None):
+    def load_lora_into_transformer(cls, state_dict, network_alphas, transformer, adapter_name=None, _pipeline=None):
         """
         This will load the LoRA layers specified in `state_dict` into `transformer`.
 
@@ -1647,6 +1668,10 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
                 A standard state dict containing the lora layer parameters. The keys can either be indexed directly
                 into the unet or prefixed with an additional `unet` which can be used to distinguish between text
                 encoder lora layers.
+            network_alphas (`Dict[str, float]`):
+                The value of the network alpha used for stable learning and preventing underflow. This value has the
+                same meaning as the `--network_alpha` option in the kohya-ss trainer script. Refer to [this
+                link](https://github.com/darkstorm2150/sd-scripts/blob/main/docs/train_network_README-en.md#execute-learning).
             transformer (`SD3Transformer2DModel`):
                 The Transformer model to load the LoRA layers into.
             adapter_name (`str`, *optional*):
@@ -1678,7 +1703,12 @@ class FluxLoraLoaderMixin(LoraBaseMixin):
                 if "lora_B" in key:
                     rank[key] = val.shape[1]
 
-            lora_config_kwargs = get_peft_kwargs(rank, network_alpha_dict=None, peft_state_dict=state_dict)
+            if network_alphas is not None and len(network_alphas) >= 1:
+                prefix = cls.transformer_name
+                alpha_keys = [k for k in network_alphas.keys() if k.startswith(prefix) and k.split(".")[0] == prefix]
+                network_alphas = {k.replace(f"{prefix}.", ""): v for k, v in network_alphas.items() if k in alpha_keys}
+
+            lora_config_kwargs = get_peft_kwargs(rank, network_alpha_dict=network_alphas, peft_state_dict=state_dict)
             if "use_dora" in lora_config_kwargs:
                 if lora_config_kwargs["use_dora"] and is_peft_version("<", "0.9.0"):
                     raise ValueError(

src/diffusers/loaders/single_file.py

@@ -23,6 +23,7 @@ from packaging import version
 from ..utils import deprecate, is_transformers_available, logging
 from .single_file_utils import (
     SingleFileComponentError,
+    _is_legacy_scheduler_kwargs,
     _is_model_weights_in_cached_folder,
     _legacy_load_clip_tokenizer,
     _legacy_load_safety_checker,
@@ -42,7 +43,6 @@ logger = logging.get_logger(__name__)
 # Legacy behaviour. `from_single_file` does not load the safety checker unless explicitly provided
 SINGLE_FILE_OPTIONAL_COMPONENTS = ["safety_checker"]
 
-
 if is_transformers_available():
     import transformers
     from transformers import PreTrainedModel, PreTrainedTokenizer
@@ -135,7 +135,7 @@ def load_single_file_sub_model(
             class_obj, checkpoint=checkpoint, config=cached_model_config_path, local_files_only=local_files_only
         )
 
-    elif is_diffusers_scheduler and is_legacy_loading:
+    elif is_diffusers_scheduler and (is_legacy_loading or _is_legacy_scheduler_kwargs(kwargs)):
         loaded_sub_model = _legacy_load_scheduler(
             class_obj, checkpoint=checkpoint, component_name=name, original_config=original_config, **kwargs
         )

src/diffusers/loaders/single_file_utils.py

@@ -79,7 +79,10 @@ CHECKPOINT_KEY_NAMES = {
     "animatediff_sdxl_beta": "up_blocks.2.motion_modules.0.temporal_transformer.norm.weight",
     "animatediff_scribble": "controlnet_cond_embedding.conv_in.weight",
     "animatediff_rgb": "controlnet_cond_embedding.weight",
-    "flux": "double_blocks.0.img_attn.norm.key_norm.scale",
+    "flux": [
+        "double_blocks.0.img_attn.norm.key_norm.scale",
+        "model.diffusion_model.double_blocks.0.img_attn.norm.key_norm.scale",
+    ],
 }
 
 DIFFUSERS_DEFAULT_PIPELINE_PATHS = {
@@ -88,11 +91,11 @@ DIFFUSERS_DEFAULT_PIPELINE_PATHS = {
     "xl_inpaint": {"pretrained_model_name_or_path": "diffusers/stable-diffusion-xl-1.0-inpainting-0.1"},
     "playground-v2-5": {"pretrained_model_name_or_path": "playgroundai/playground-v2.5-1024px-aesthetic"},
     "upscale": {"pretrained_model_name_or_path": "stabilityai/stable-diffusion-x4-upscaler"},
-    "inpainting": {"pretrained_model_name_or_path": "runwayml/stable-diffusion-inpainting"},
+    "inpainting": {"pretrained_model_name_or_path": "Lykon/dreamshaper-8-inpainting"},
     "inpainting_v2": {"pretrained_model_name_or_path": "stabilityai/stable-diffusion-2-inpainting"},
     "controlnet": {"pretrained_model_name_or_path": "lllyasviel/control_v11p_sd15_canny"},
     "v2": {"pretrained_model_name_or_path": "stabilityai/stable-diffusion-2-1"},
-    "v1": {"pretrained_model_name_or_path": "runwayml/stable-diffusion-v1-5"},
+    "v1": {"pretrained_model_name_or_path": "Lykon/dreamshaper-8"},
     "stable_cascade_stage_b": {"pretrained_model_name_or_path": "stabilityai/stable-cascade", "subfolder": "decoder"},
     "stable_cascade_stage_b_lite": {
         "pretrained_model_name_or_path": "stabilityai/stable-cascade",
@@ -258,7 +261,7 @@ SCHEDULER_DEFAULT_CONFIG = {
     "timestep_spacing": "leading",
 }
 
-LDM_VAE_KEY = "first_stage_model."
+LDM_VAE_KEYS = ["first_stage_model.", "vae."]
 LDM_VAE_DEFAULT_SCALING_FACTOR = 0.18215
 PLAYGROUND_VAE_SCALING_FACTOR = 0.5
 LDM_UNET_KEY = "model.diffusion_model."
@@ -267,8 +270,8 @@ LDM_CLIP_PREFIX_TO_REMOVE = [
     "cond_stage_model.transformer.",
     "conditioner.embedders.0.transformer.",
 ]
-OPEN_CLIP_PREFIX = "conditioner.embedders.0.model."
 LDM_OPEN_CLIP_TEXT_PROJECTION_DIM = 1024
+SCHEDULER_LEGACY_KWARGS = ["prediction_type", "scheduler_type"]
 
 VALID_URL_PREFIXES = ["https://huggingface.co/", "huggingface.co/", "hf.co/", "https://hf.co/"]
 
@@ -318,6 +321,10 @@ def _is_model_weights_in_cached_folder(cached_folder, name):
     return weights_exist
 
 
+def _is_legacy_scheduler_kwargs(kwargs):
+    return any(k in SCHEDULER_LEGACY_KWARGS for k in kwargs.keys())
+
+
 def load_single_file_checkpoint(
     pretrained_model_link_or_path,
     force_download=False,
@@ -516,8 +523,10 @@ def infer_diffusers_model_type(checkpoint):
         else:
             model_type = "animatediff_v3"
 
-    elif CHECKPOINT_KEY_NAMES["flux"] in checkpoint:
-        if "guidance_in.in_layer.bias" in checkpoint:
+    elif any(key in checkpoint for key in CHECKPOINT_KEY_NAMES["flux"]):
+        if any(
+            g in checkpoint for g in ["guidance_in.in_layer.bias", "model.diffusion_model.guidance_in.in_layer.bias"]
+        ):
             model_type = "flux-dev"
         else:
             model_type = "flux-schnell"
@@ -1176,7 +1185,11 @@ def convert_ldm_vae_checkpoint(checkpoint, config):
     # remove the LDM_VAE_KEY prefix from the ldm checkpoint keys so that it is easier to map them to diffusers keys
     vae_state_dict = {}
     keys = list(checkpoint.keys())
-    vae_key = LDM_VAE_KEY if any(k.startswith(LDM_VAE_KEY) for k in keys) else ""
+    vae_key = ""
+    for ldm_vae_key in LDM_VAE_KEYS:
+        if any(k.startswith(ldm_vae_key) for k in keys):
+            vae_key = ldm_vae_key
+
     for key in keys:
         if key.startswith(vae_key):
             vae_state_dict[key.replace(vae_key, "")] = checkpoint.get(key)
@@ -1477,14 +1490,22 @@ def _legacy_load_scheduler(
 
     if scheduler_type is not None:
         deprecation_message = (
-            "Please pass an instance of a Scheduler object directly to the `scheduler` argument in `from_single_file`."
+            "Please pass an instance of a Scheduler object directly to the `scheduler` argument in `from_single_file`\n\n"
+            "Example:\n\n"
+            "from diffusers import StableDiffusionPipeline, DDIMScheduler\n\n"
+            "scheduler = DDIMScheduler()\n"
+            "pipe = StableDiffusionPipeline.from_single_file(<checkpoint path>, scheduler=scheduler)\n"
         )
         deprecate("scheduler_type", "1.0.0", deprecation_message)
 
     if prediction_type is not None:
         deprecation_message = (
-            "Please configure an instance of a Scheduler with the appropriate `prediction_type` "
-            "and pass the object directly to the `scheduler` argument in `from_single_file`."
+            "Please configure an instance of a Scheduler with the appropriate `prediction_type` and "
+            "pass the object directly to the `scheduler` argument in `from_single_file`.\n\n"
+            "Example:\n\n"
+            "from diffusers import StableDiffusionPipeline, DDIMScheduler\n\n"
+            'scheduler = DDIMScheduler(prediction_type="v_prediction")\n'
+            "pipe = StableDiffusionPipeline.from_single_file(<checkpoint path>, scheduler=scheduler)\n"
         )
         deprecate("prediction_type", "1.0.0", deprecation_message)
 
@@ -1881,6 +1902,10 @@ def convert_animatediff_checkpoint_to_diffusers(checkpoint, **kwargs):
 
 def convert_flux_transformer_checkpoint_to_diffusers(checkpoint, **kwargs):
     converted_state_dict = {}
+    keys = list(checkpoint.keys())
+    for k in keys:
+        if "model.diffusion_model." in k:
+            checkpoint[k.replace("model.diffusion_model.", "")] = checkpoint.pop(k)
 
     num_layers = list(set(int(k.split(".", 2)[1]) for k in checkpoint if "double_blocks." in k))[-1] + 1  # noqa: C401
     num_single_layers = list(set(int(k.split(".", 2)[1]) for k in checkpoint if "single_blocks." in k))[-1] + 1  # noqa: C401

src/diffusers/models/__init__.py

@@ -35,7 +35,6 @@ if is_torch_available():
     _import_structure["autoencoders.consistency_decoder_vae"] = ["ConsistencyDecoderVAE"]
     _import_structure["autoencoders.vq_model"] = ["VQModel"]
     _import_structure["controlnet"] = ["ControlNetModel"]
-    _import_structure["controlnet_flux"] = ["FluxControlNetModel"]
     _import_structure["controlnet_hunyuan"] = ["HunyuanDiT2DControlNetModel", "HunyuanDiT2DMultiControlNetModel"]
     _import_structure["controlnet_sd3"] = ["SD3ControlNetModel", "SD3MultiControlNetModel"]
     _import_structure["controlnet_sparsectrl"] = ["SparseControlNetModel"]
@@ -88,7 +87,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             VQModel,
         )
         from .controlnet import ControlNetModel
-        from .controlnet_flux import FluxControlNetModel
         from .controlnet_hunyuan import HunyuanDiT2DControlNetModel, HunyuanDiT2DMultiControlNetModel
         from .controlnet_sd3 import SD3ControlNetModel, SD3MultiControlNetModel
         from .controlnet_sparsectrl import SparseControlNetModel

src/diffusers/models/attention.py

@@ -449,7 +449,7 @@ class BasicTransformerBlock(nn.Module):
             norm_hidden_states = self.norm1(hidden_states, added_cond_kwargs["pooled_text_emb"])
         elif self.norm_type == "ada_norm_single":
             shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
-                self.scale_shift_table[None].to(timestep.dtype) + timestep.reshape(batch_size, 6, -1)
+                self.scale_shift_table[None] + timestep.reshape(batch_size, 6, -1)
             ).chunk(6, dim=1)
             norm_hidden_states = self.norm1(hidden_states)
             norm_hidden_states = norm_hidden_states * (1 + scale_msa) + shift_msa

src/diffusers/models/attention_processor.py

@@ -1868,6 +1868,148 @@ class FluxAttnProcessor2_0:
         return hidden_states, encoder_hidden_states
 
 
+class CogVideoXAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention for the CogVideoX model. It applies a rotary embedding on
+    query and key vectors, but does not include spatial normalization.
+    """
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("CogVideoXAttnProcessor requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+    def __call__(
+        self,
+        attn: Attention,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        image_rotary_emb: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        text_seq_length = encoder_hidden_states.size(1)
+
+        hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        query = attn.to_q(hidden_states)
+        key = attn.to_k(hidden_states)
+        value = attn.to_v(hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        if attn.norm_q is not None:
+            query = attn.norm_q(query)
+        if attn.norm_k is not None:
+            key = attn.norm_k(key)
+
+        # Apply RoPE if needed
+        if image_rotary_emb is not None:
+            from .embeddings import apply_rotary_emb
+
+            query[:, :, text_seq_length:] = apply_rotary_emb(query[:, :, text_seq_length:], image_rotary_emb)
+            if not attn.is_cross_attention:
+                key[:, :, text_seq_length:] = apply_rotary_emb(key[:, :, text_seq_length:], image_rotary_emb)
+
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        encoder_hidden_states, hidden_states = hidden_states.split(
+            [text_seq_length, hidden_states.size(1) - text_seq_length], dim=1
+        )
+        return hidden_states, encoder_hidden_states
+
+
+class FusedCogVideoXAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention for the CogVideoX model. It applies a rotary embedding on
+    query and key vectors, but does not include spatial normalization.
+    """
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("CogVideoXAttnProcessor requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+    def __call__(
+        self,
+        attn: Attention,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        image_rotary_emb: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        text_seq_length = encoder_hidden_states.size(1)
+
+        hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        qkv = attn.to_qkv(hidden_states)
+        split_size = qkv.shape[-1] // 3
+        query, key, value = torch.split(qkv, split_size, dim=-1)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        if attn.norm_q is not None:
+            query = attn.norm_q(query)
+        if attn.norm_k is not None:
+            key = attn.norm_k(key)
+
+        # Apply RoPE if needed
+        if image_rotary_emb is not None:
+            from .embeddings import apply_rotary_emb
+
+            query[:, :, text_seq_length:] = apply_rotary_emb(query[:, :, text_seq_length:], image_rotary_emb)
+            if not attn.is_cross_attention:
+                key[:, :, text_seq_length:] = apply_rotary_emb(key[:, :, text_seq_length:], image_rotary_emb)
+
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        encoder_hidden_states, hidden_states = hidden_states.split(
+            [text_seq_length, hidden_states.size(1) - text_seq_length], dim=1
+        )
+        return hidden_states, encoder_hidden_states
+
+
 class XFormersAttnAddedKVProcessor:
     r"""
     Processor for implementing memory efficient attention using xFormers.

src/diffusers/models/autoencoders/autoencoder_asym_kl.py

@@ -60,8 +60,6 @@ class AsymmetricAutoencoderKL(ModelMixin, ConfigMixin):
             Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) paper.
     """
 
-    _always_upcast_modules = ["MaskConditionDecoder"]
-
     @register_to_config
     def __init__(
         self,

src/diffusers/models/autoencoders/autoencoder_kl.py

@@ -70,7 +70,6 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin):
 
     _supports_gradient_checkpointing = True
     _no_split_modules = ["BasicTransformerBlock", "ResnetBlock2D"]
-    _always_upcast_modules = ["Decoder"]
 
     @register_to_config
     def __init__(

src/diffusers/models/autoencoders/autoencoder_kl_cogvideox.py

@@ -902,7 +902,7 @@ class AutoencoderKLCogVideoX(ModelMixin, ConfigMixin, FromOriginalModelMixin):
             Tuple of block output channels.
         act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use.
         sample_size (`int`, *optional*, defaults to `32`): Sample input size.
-        scaling_factor (`float`, *optional*, defaults to 0.18215):
+        scaling_factor (`float`, *optional*, defaults to `1.15258426`):
             The component-wise standard deviation of the trained latent space computed using the first batch of the
             training set. This is used to scale the latent space to have unit variance when training the diffusion
             model. The latents are scaled with the formula `z = z * scaling_factor` before being passed to the

src/diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py

@@ -192,7 +192,6 @@ class AutoencoderKLTemporalDecoder(ModelMixin, ConfigMixin):
     """
 
     _supports_gradient_checkpointing = True
-    _always_upcast_modules = ["TemporalDecoder"]
 
     @register_to_config
     def __init__(

src/diffusers/models/autoencoders/autoencoder_oobleck.py

@@ -317,7 +317,6 @@ class AutoencoderOobleck(ModelMixin, ConfigMixin):
     """
 
     _supports_gradient_checkpointing = False
-    _always_upcast_modules = ["OobleckEncoder", "OobleckDecoder"]
 
     @register_to_config
     def __init__(

src/diffusers/models/autoencoders/consistency_decoder_vae.py

@@ -330,7 +330,7 @@ class ConsistencyDecoderVAE(ModelMixin, ConfigMixin):
             Union[DecoderOutput, Tuple[torch.Tensor]]: The decoded output.
 
         """
-        z = (z * self.config.scaling_factor - self.means.to(z.dtype)) / self.stds.to(z.dtype)
+        z = (z * self.config.scaling_factor - self.means) / self.stds
 
         scale_factor = 2 ** (len(self.config.block_out_channels) - 1)
         z = F.interpolate(z, mode="nearest", scale_factor=scale_factor)

src/diffusers/models/autoencoders/vq_model.py

@@ -71,8 +71,6 @@ class VQModel(ModelMixin, ConfigMixin):
             Type of normalization layer to use. Can be one of `"group"` or `"spatial"`.
     """
 
-    _always_upcast_modules = ["Decoder", "VectorQuantizer"]
-
     @register_to_config
     def __init__(
         self,

src/diffusers/models/controlnet_flux.py

@@ -1,374 +0,0 @@
-# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional, Tuple, Union
-
-import torch
-import torch.nn as nn
-
-from ..configuration_utils import ConfigMixin, register_to_config
-from ..loaders import PeftAdapterMixin
-from ..models.attention_processor import AttentionProcessor
-from ..models.modeling_utils import ModelMixin
-from ..utils import USE_PEFT_BACKEND, is_torch_version, logging, scale_lora_layers, unscale_lora_layers
-from .controlnet import BaseOutput, zero_module
-from .embeddings import CombinedTimestepGuidanceTextProjEmbeddings, CombinedTimestepTextProjEmbeddings
-from .modeling_outputs import Transformer2DModelOutput
-from .transformers.transformer_flux import EmbedND, FluxSingleTransformerBlock, FluxTransformerBlock
-
-
-logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
-
-
-@dataclass
-class FluxControlNetOutput(BaseOutput):
-    controlnet_block_samples: Tuple[torch.Tensor]
-    controlnet_single_block_samples: Tuple[torch.Tensor]
-
-
-class FluxControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
-    _supports_gradient_checkpointing = True
-
-    @register_to_config
-    def __init__(
-        self,
-        patch_size: int = 1,
-        in_channels: int = 64,
-        num_layers: int = 19,
-        num_single_layers: int = 38,
-        attention_head_dim: int = 128,
-        num_attention_heads: int = 24,
-        joint_attention_dim: int = 4096,
-        pooled_projection_dim: int = 768,
-        guidance_embeds: bool = False,
-        axes_dims_rope: List[int] = [16, 56, 56],
-    ):
-        super().__init__()
-        self.out_channels = in_channels
-        self.inner_dim = num_attention_heads * attention_head_dim
-
-        self.pos_embed = EmbedND(dim=self.inner_dim, theta=10000, axes_dim=axes_dims_rope)
-        text_time_guidance_cls = (
-            CombinedTimestepGuidanceTextProjEmbeddings if guidance_embeds else CombinedTimestepTextProjEmbeddings
-        )
-        self.time_text_embed = text_time_guidance_cls(
-            embedding_dim=self.inner_dim, pooled_projection_dim=pooled_projection_dim
-        )
-
-        self.context_embedder = nn.Linear(joint_attention_dim, self.inner_dim)
-        self.x_embedder = torch.nn.Linear(in_channels, self.inner_dim)
-
-        self.transformer_blocks = nn.ModuleList(
-            [
-                FluxTransformerBlock(
-                    dim=self.inner_dim,
-                    num_attention_heads=num_attention_heads,
-                    attention_head_dim=attention_head_dim,
-                )
-                for i in range(num_layers)
-            ]
-        )
-
-        self.single_transformer_blocks = nn.ModuleList(
-            [
-                FluxSingleTransformerBlock(
-                    dim=self.inner_dim,
-                    num_attention_heads=num_attention_heads,
-                    attention_head_dim=attention_head_dim,
-                )
-                for i in range(num_single_layers)
-            ]
-        )
-
-        # controlnet_blocks
-        self.controlnet_blocks = nn.ModuleList([])
-        for _ in range(len(self.transformer_blocks)):
-            self.controlnet_blocks.append(zero_module(nn.Linear(self.inner_dim, self.inner_dim)))
-
-        self.controlnet_single_blocks = nn.ModuleList([])
-        for _ in range(len(self.single_transformer_blocks)):
-            self.controlnet_single_blocks.append(zero_module(nn.Linear(self.inner_dim, self.inner_dim)))
-
-        self.controlnet_x_embedder = zero_module(torch.nn.Linear(in_channels, self.inner_dim))
-
-        self.gradient_checkpointing = False
-
-    @property
-    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
-    def attn_processors(self):
-        r"""
-        Returns:
-            `dict` of attention processors: A dictionary containing all attention processors used in the model with
-            indexed by its weight name.
-        """
-        # set recursively
-        processors = {}
-
-        def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
-            if hasattr(module, "get_processor"):
-                processors[f"{name}.processor"] = module.get_processor()
-
-            for sub_name, child in module.named_children():
-                fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
-
-            return processors
-
-        for name, module in self.named_children():
-            fn_recursive_add_processors(name, module, processors)
-
-        return processors
-
-    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
-    def set_attn_processor(self, processor):
-        r"""
-        Sets the attention processor to use to compute attention.
-
-        Parameters:
-            processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
-                The instantiated processor class or a dictionary of processor classes that will be set as the processor
-                for **all** `Attention` layers.
-
-                If `processor` is a dict, the key needs to define the path to the corresponding cross attention
-                processor. This is strongly recommended when setting trainable attention processors.
-
-        """
-        count = len(self.attn_processors.keys())
-
-        if isinstance(processor, dict) and len(processor) != count:
-            raise ValueError(
-                f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
-                f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
-            )
-
-        def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
-            if hasattr(module, "set_processor"):
-                if not isinstance(processor, dict):
-                    module.set_processor(processor)
-                else:
-                    module.set_processor(processor.pop(f"{name}.processor"))
-
-            for sub_name, child in module.named_children():
-                fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
-
-        for name, module in self.named_children():
-            fn_recursive_attn_processor(name, module, processor)
-
-    def _set_gradient_checkpointing(self, module, value=False):
-        if hasattr(module, "gradient_checkpointing"):
-            module.gradient_checkpointing = value
-
-    @classmethod
-    def from_transformer(
-        cls,
-        transformer,
-        num_layers=4,
-        num_single_layers=10,
-        attention_head_dim: int = 128,
-        num_attention_heads: int = 24,
-        load_weights_from_transformer=True,
-    ):
-        config = transformer.config
-        config["num_layers"] = num_layers
-        config["num_single_layers"] = num_single_layers
-        config["attention_head_dim"] = attention_head_dim
-        config["num_attention_heads"] = num_attention_heads
-
-        controlnet = cls(**config)
-
-        if load_weights_from_transformer:
-            controlnet.pos_embed.load_state_dict(transformer.pos_embed.state_dict())
-            controlnet.time_text_embed.load_state_dict(transformer.time_text_embed.state_dict())
-            controlnet.context_embedder.load_state_dict(transformer.context_embedder.state_dict())
-            controlnet.x_embedder.load_state_dict(transformer.x_embedder.state_dict())
-            controlnet.transformer_blocks.load_state_dict(transformer.transformer_blocks.state_dict(), strict=False)
-            controlnet.single_transformer_blocks.load_state_dict(
-                transformer.single_transformer_blocks.state_dict(), strict=False
-            )
-
-            controlnet.controlnet_x_embedder = zero_module(controlnet.controlnet_x_embedder)
-
-        return controlnet
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        controlnet_cond: torch.Tensor,
-        conditioning_scale: float = 1.0,
-        encoder_hidden_states: torch.Tensor = None,
-        pooled_projections: torch.Tensor = None,
-        timestep: torch.LongTensor = None,
-        img_ids: torch.Tensor = None,
-        txt_ids: torch.Tensor = None,
-        guidance: torch.Tensor = None,
-        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
-        return_dict: bool = True,
-    ) -> Union[torch.FloatTensor, Transformer2DModelOutput]:
-        """
-        The [`FluxTransformer2DModel`] forward method.
-
-        Args:
-            hidden_states (`torch.FloatTensor` of shape `(batch size, channel, height, width)`):
-                Input `hidden_states`.
-            encoder_hidden_states (`torch.FloatTensor` of shape `(batch size, sequence_len, embed_dims)`):
-                Conditional embeddings (embeddings computed from the input conditions such as prompts) to use.
-            pooled_projections (`torch.FloatTensor` of shape `(batch_size, projection_dim)`): Embeddings projected
-                from the embeddings of input conditions.
-            timestep ( `torch.LongTensor`):
-                Used to indicate denoising step.
-            block_controlnet_hidden_states: (`list` of `torch.Tensor`):
-                A list of tensors that if specified are added to the residuals of transformer blocks.
-            joint_attention_kwargs (`dict`, *optional*):
-                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
-                `self.processor` in
-                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
-            return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain
-                tuple.
-
-        Returns:
-            If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
-            `tuple` where the first element is the sample tensor.
-        """
-        if joint_attention_kwargs is not None:
-            joint_attention_kwargs = joint_attention_kwargs.copy()
-            lora_scale = joint_attention_kwargs.pop("scale", 1.0)
-        else:
-            lora_scale = 1.0
-
-        if USE_PEFT_BACKEND:
-            # weight the lora layers by setting `lora_scale` for each PEFT layer
-            scale_lora_layers(self, lora_scale)
-        else:
-            if joint_attention_kwargs is not None and joint_attention_kwargs.get("scale", None) is not None:
-                logger.warning(
-                    "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective."
-                )
-        hidden_states = self.x_embedder(hidden_states)
-
-        # add
-        hidden_states = hidden_states + self.controlnet_x_embedder(controlnet_cond)
-
-        timestep = timestep.to(hidden_states.dtype) * 1000
-        if guidance is not None:
-            guidance = guidance.to(hidden_states.dtype) * 1000
-        else:
-            guidance = None
-        temb = (
-            self.time_text_embed(timestep, pooled_projections)
-            if guidance is None
-            else self.time_text_embed(timestep, guidance, pooled_projections)
-        )
-        encoder_hidden_states = self.context_embedder(encoder_hidden_states)
-
-        txt_ids = txt_ids.expand(img_ids.size(0), -1, -1)
-        ids = torch.cat((txt_ids, img_ids), dim=1)
-        image_rotary_emb = self.pos_embed(ids)
-
-        block_samples = ()
-        for index_block, block in enumerate(self.transformer_blocks):
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module, return_dict=None):
-                    def custom_forward(*inputs):
-                        if return_dict is not None:
-                            return module(*inputs, return_dict=return_dict)
-                        else:
-                            return module(*inputs)
-
-                    return custom_forward
-
-                ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
-                encoder_hidden_states, hidden_states = torch.utils.checkpoint.checkpoint(
-                    create_custom_forward(block),
-                    hidden_states,
-                    encoder_hidden_states,
-                    temb,
-                    image_rotary_emb,
-                    **ckpt_kwargs,
-                )
-
-            else:
-                encoder_hidden_states, hidden_states = block(
-                    hidden_states=hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    temb=temb,
-                    image_rotary_emb=image_rotary_emb,
-                )
-            block_samples = block_samples + (hidden_states,)
-
-        hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
-
-        single_block_samples = ()
-        for index_block, block in enumerate(self.single_transformer_blocks):
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module, return_dict=None):
-                    def custom_forward(*inputs):
-                        if return_dict is not None:
-                            return module(*inputs, return_dict=return_dict)
-                        else:
-                            return module(*inputs)
-
-                    return custom_forward
-
-                ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
-                hidden_states = torch.utils.checkpoint.checkpoint(
-                    create_custom_forward(block),
-                    hidden_states,
-                    temb,
-                    image_rotary_emb,
-                    **ckpt_kwargs,
-                )
-
-            else:
-                hidden_states = block(
-                    hidden_states=hidden_states,
-                    temb=temb,
-                    image_rotary_emb=image_rotary_emb,
-                )
-            single_block_samples = single_block_samples + (hidden_states[:, encoder_hidden_states.shape[1] :],)
-
-        # controlnet block
-        controlnet_block_samples = ()
-        for block_sample, controlnet_block in zip(block_samples, self.controlnet_blocks):
-            block_sample = controlnet_block(block_sample)
-            controlnet_block_samples = controlnet_block_samples + (block_sample,)
-
-        controlnet_single_block_samples = ()
-        for single_block_sample, controlnet_block in zip(single_block_samples, self.controlnet_single_blocks):
-            single_block_sample = controlnet_block(single_block_sample)
-            controlnet_single_block_samples = controlnet_single_block_samples + (single_block_sample,)
-
-        # scaling
-        controlnet_block_samples = [sample * conditioning_scale for sample in controlnet_block_samples]
-        controlnet_single_block_samples = [sample * conditioning_scale for sample in controlnet_single_block_samples]
-
-        #
-        controlnet_block_samples = None if len(controlnet_block_samples) == 0 else controlnet_block_samples
-        controlnet_single_block_samples = (
-            None if len(controlnet_single_block_samples) == 0 else controlnet_single_block_samples
-        )
-
-        if USE_PEFT_BACKEND:
-            # remove `lora_scale` from each PEFT layer
-            unscale_lora_layers(self, lora_scale)
-
-        if not return_dict:
-            return (controlnet_block_samples, controlnet_single_block_samples)
-
-        return FluxControlNetOutput(
-            controlnet_block_samples=controlnet_block_samples,
-            controlnet_single_block_samples=controlnet_single_block_samples,
-        )

src/diffusers/models/controlnet_sd3.py

@@ -55,7 +55,6 @@ class SD3ControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginal
         pooled_projection_dim: int = 2048,
         out_channels: int = 16,
         pos_embed_max_size: int = 96,
-        extra_conditioning_channels: int = 0,
     ):
         super().__init__()
         default_out_channels = in_channels
@@ -99,7 +98,7 @@ class SD3ControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginal
             height=sample_size,
             width=sample_size,
             patch_size=patch_size,
-            in_channels=in_channels + extra_conditioning_channels,
+            in_channels=in_channels,
             embed_dim=self.inner_dim,
             pos_embed_type=None,
         )

src/diffusers/models/embeddings.py

@@ -374,6 +374,90 @@ class CogVideoXPatchEmbed(nn.Module):
         return embeds
 
 
+def get_3d_rotary_pos_embed(
+    embed_dim, crops_coords, grid_size, temporal_size, theta: int = 10000, use_real: bool = True
+) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
+    """
+    RoPE for video tokens with 3D structure.
+
+    Args:
+    embed_dim: (`int`):
+        The embedding dimension size, corresponding to hidden_size_head.
+    crops_coords (`Tuple[int]`):
+        The top-left and bottom-right coordinates of the crop.
+    grid_size (`Tuple[int]`):
+        The grid size of the spatial positional embedding (height, width).
+    temporal_size (`int`):
+        The size of the temporal dimension.
+    theta (`float`):
+        Scaling factor for frequency computation.
+    use_real (`bool`):
+        If True, return real part and imaginary part separately. Otherwise, return complex numbers.
+
+    Returns:
+        `torch.Tensor`: positional embedding with shape `(temporal_size * grid_size[0] * grid_size[1], embed_dim/2)`.
+    """
+    start, stop = crops_coords
+    grid_h = np.linspace(start[0], stop[0], grid_size[0], endpoint=False, dtype=np.float32)
+    grid_w = np.linspace(start[1], stop[1], grid_size[1], endpoint=False, dtype=np.float32)
+    grid_t = np.linspace(0, temporal_size, temporal_size, endpoint=False, dtype=np.float32)
+
+    # Compute dimensions for each axis
+    dim_t = embed_dim // 4
+    dim_h = embed_dim // 8 * 3
+    dim_w = embed_dim // 8 * 3
+
+    # Temporal frequencies
+    freqs_t = 1.0 / (theta ** (torch.arange(0, dim_t, 2).float() / dim_t))
+    grid_t = torch.from_numpy(grid_t).float()
+    freqs_t = torch.einsum("n , f -> n f", grid_t, freqs_t)
+    freqs_t = freqs_t.repeat_interleave(2, dim=-1)
+
+    # Spatial frequencies for height and width
+    freqs_h = 1.0 / (theta ** (torch.arange(0, dim_h, 2).float() / dim_h))
+    freqs_w = 1.0 / (theta ** (torch.arange(0, dim_w, 2).float() / dim_w))
+    grid_h = torch.from_numpy(grid_h).float()
+    grid_w = torch.from_numpy(grid_w).float()
+    freqs_h = torch.einsum("n , f -> n f", grid_h, freqs_h)
+    freqs_w = torch.einsum("n , f -> n f", grid_w, freqs_w)
+    freqs_h = freqs_h.repeat_interleave(2, dim=-1)
+    freqs_w = freqs_w.repeat_interleave(2, dim=-1)
+
+    # Broadcast and concatenate tensors along specified dimension
+    def broadcast(tensors, dim=-1):
+        num_tensors = len(tensors)
+        shape_lens = {len(t.shape) for t in tensors}
+        assert len(shape_lens) == 1, "tensors must all have the same number of dimensions"
+        shape_len = list(shape_lens)[0]
+        dim = (dim + shape_len) if dim < 0 else dim
+        dims = list(zip(*(list(t.shape) for t in tensors)))
+        expandable_dims = [(i, val) for i, val in enumerate(dims) if i != dim]
+        assert all(
+            [*(len(set(t[1])) <= 2 for t in expandable_dims)]
+        ), "invalid dimensions for broadcastable concatenation"
+        max_dims = [(t[0], max(t[1])) for t in expandable_dims]
+        expanded_dims = [(t[0], (t[1],) * num_tensors) for t in max_dims]
+        expanded_dims.insert(dim, (dim, dims[dim]))
+        expandable_shapes = list(zip(*(t[1] for t in expanded_dims)))
+        tensors = [t[0].expand(*t[1]) for t in zip(tensors, expandable_shapes)]
+        return torch.cat(tensors, dim=dim)
+
+    freqs = broadcast((freqs_t[:, None, None, :], freqs_h[None, :, None, :], freqs_w[None, None, :, :]), dim=-1)
+
+    t, h, w, d = freqs.shape
+    freqs = freqs.view(t * h * w, d)
+
+    # Generate sine and cosine components
+    sin = freqs.sin()
+    cos = freqs.cos()
+
+    if use_real:
+        return cos, sin
+    else:
+        freqs_cis = torch.polar(torch.ones_like(freqs), freqs)
+        return freqs_cis
+
+
 def get_2d_rotary_pos_embed(embed_dim, crops_coords, grid_size, use_real=True):
     """
     RoPE for image tokens with 2d structure.

src/diffusers/models/modeling_utils.py

@@ -263,80 +263,6 @@ class ModelMixin(torch.nn.Module, PushToHubMixin):
         """
         self.set_use_memory_efficient_attention_xformers(False)
 
-    def enable_layerwise_upcasting(self, upcast_dtype=None):
-        r"""
-        Enable layerwise dynamic upcasting. This allows models to be loaded into the GPU in a low memory dtype e.g.
-        torch.float8_e4m3fn, but perform inference using a dtype that is supported by the GPU, by upcasting the
-        individual modules in the model to the appropriate dtype right before the foward pass.
-
-        The module is then moved back to the low memory dtype after the foward pass.
-        """
-
-        upcast_dtype = upcast_dtype or torch.float32
-        original_dtype = self.dtype
-
-        def upcast_dtype_hook_fn(module, *args, **kwargs):
-            module = module.to(upcast_dtype)
-
-        def cast_to_original_dtype_hook_fn(module, *args, **kwargs):
-            module = module.to(original_dtype)
-
-        def fn_recursive_upcast(module):
-            """In certain cases modules will apply casting internally or reference the dtype of internal blocks.
-
-            e.g.
-
-            ```
-            class MyModel(nn.Module):
-                def forward(self, x):
-                    dtype = next(iter(self.blocks.parameters())).dtype
-                    x = self.blocks(x) + torch.ones(x.size()).to(dtype)
-            ```
-            Layerwise upcasting will not work here, since the internal blocks remain in the low memory dtype until
-            their `forward` method is called. We need to add the upcast hook on the entire module in order for the
-            operation to work.
-
-            The `_always_upcast_modules` class attribute is a list of modules within the model that we must upcast
-            entirely, rather than layerwise.
-
-            """
-            if hasattr(self, "_always_upcast_modules") and module.__class__.__name__ in self._always_upcast_modules:
-                # Upcast entire module and exist recursion
-                module.register_forward_pre_hook(upcast_dtype_hook_fn)
-                module.register_forward_hook(cast_to_original_dtype_hook_fn)
-
-                return
-
-            has_children = list(module.children())
-            if not has_children:
-                module.register_forward_pre_hook(upcast_dtype_hook_fn)
-                module.register_forward_hook(cast_to_original_dtype_hook_fn)
-
-            for child in module.children():
-                fn_recursive_upcast(child)
-
-        for module in self.children():
-            fn_recursive_upcast(module)
-
-    def disable_layerwise_upcasting(self):
-        def fn_recursive_upcast(module):
-            if hasattr(self, "_always_upcast_modules") and module.__class__.__name__ in self._always_upcast_modules:
-                module._forward_pre_hooks = OrderedDict()
-                module._forward_hooks = OrderedDict()
-
-                return
-
-            has_children = list(module.children())
-            if not has_children:
-                module._forward_pre_hooks = OrderedDict()
-                module._forward_hooks = OrderedDict()
-
-            for child in module.children():
-                fn_recursive_upcast(child)
-
-        for module in self.children():
-            fn_recursive_upcast(module)
-
     def save_pretrained(
         self,
         save_directory: Union[str, os.PathLike],

src/diffusers/models/transformers/auraflow_transformer_2d.py

@@ -274,9 +274,7 @@ class AuraFlowTransformer2DModel(ModelMixin, ConfigMixin):
         pos_embed_max_size (`int`, defaults to 4096): Maximum positions to embed from the image latents.
     """
 
-    _no_split_modules = ["AuraFlowJointTransformerBlock", "AuraFlowSingleTransformerBlock", "AuraFlowPatchEmbed"]
     _supports_gradient_checkpointing = True
-    _always_upcast_modules = ["AuraFlowPatchEmbed"]
 
     @register_to_config
     def __init__(
@@ -458,15 +456,11 @@ class AuraFlowTransformer2DModel(ModelMixin, ConfigMixin):
 
         # Apply patch embedding, timestep embedding, and project the caption embeddings.
         hidden_states = self.pos_embed(hidden_states)  # takes care of adding positional embeddings too.
-        temb = self.time_step_embed(timestep).to(dtype=hidden_states.dtype)
+        temb = self.time_step_embed(timestep).to(dtype=next(self.parameters()).dtype)
         temb = self.time_step_proj(temb)
         encoder_hidden_states = self.context_embedder(encoder_hidden_states)
         encoder_hidden_states = torch.cat(
-            [
-                self.register_tokens.to(encoder_hidden_states.dtype).repeat(encoder_hidden_states.size(0), 1, 1),
-                encoder_hidden_states,
-            ],
-            dim=1,
+            [self.register_tokens.repeat(encoder_hidden_states.size(0), 1, 1), encoder_hidden_states], dim=1
         )
 
         # MMDiT blocks.

src/diffusers/models/transformers/cogvideox_transformer_3d.py

@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any, Dict, Optional, Union
+from typing import Any, Dict, Optional, Tuple, Union
 
 import torch
 from torch import nn
@@ -22,6 +22,7 @@ from ...configuration_utils import ConfigMixin, register_to_config
 from ...utils import is_torch_version, logging
 from ...utils.torch_utils import maybe_allow_in_graph
 from ..attention import Attention, FeedForward
+from ..attention_processor import AttentionProcessor, CogVideoXAttnProcessor2_0, FusedCogVideoXAttnProcessor2_0
 from ..embeddings import CogVideoXPatchEmbed, TimestepEmbedding, Timesteps, get_3d_sincos_pos_embed
 from ..modeling_outputs import Transformer2DModelOutput
 from ..modeling_utils import ModelMixin
@@ -97,6 +98,7 @@ class CogVideoXBlock(nn.Module):
             eps=1e-6,
             bias=attention_bias,
             out_bias=attention_out_bias,
+            processor=CogVideoXAttnProcessor2_0(),
         )
 
         # 2. Feed Forward
@@ -116,24 +118,24 @@ class CogVideoXBlock(nn.Module):
         hidden_states: torch.Tensor,
         encoder_hidden_states: torch.Tensor,
         temb: torch.Tensor,
+        image_rotary_emb: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
     ) -> torch.Tensor:
+        text_seq_length = encoder_hidden_states.size(1)
+
+        # norm & modulate
         norm_hidden_states, norm_encoder_hidden_states, gate_msa, enc_gate_msa = self.norm1(
             hidden_states, encoder_hidden_states, temb
         )
 
         # attention
-        text_length = norm_encoder_hidden_states.size(1)
-
-        # CogVideoX uses concatenated text + video embeddings with self-attention instead of using
-        # them in cross-attention individually
-        norm_hidden_states = torch.cat([norm_encoder_hidden_states, norm_hidden_states], dim=1)
-        attn_output = self.attn1(
+        attn_hidden_states, attn_encoder_hidden_states = self.attn1(
             hidden_states=norm_hidden_states,
-            encoder_hidden_states=None,
+            encoder_hidden_states=norm_encoder_hidden_states,
+            image_rotary_emb=image_rotary_emb,
         )
 
-        hidden_states = hidden_states + gate_msa * attn_output[:, text_length:]
-        encoder_hidden_states = encoder_hidden_states + enc_gate_msa * attn_output[:, :text_length]
+        hidden_states = hidden_states + gate_msa * attn_hidden_states
+        encoder_hidden_states = encoder_hidden_states + enc_gate_msa * attn_encoder_hidden_states
 
         # norm & modulate
         norm_hidden_states, norm_encoder_hidden_states, gate_ff, enc_gate_ff = self.norm2(
@@ -144,8 +146,9 @@ class CogVideoXBlock(nn.Module):
         norm_hidden_states = torch.cat([norm_encoder_hidden_states, norm_hidden_states], dim=1)
         ff_output = self.ff(norm_hidden_states)
 
-        hidden_states = hidden_states + gate_ff * ff_output[:, text_length:]
-        encoder_hidden_states = encoder_hidden_states + enc_gate_ff * ff_output[:, :text_length]
+        hidden_states = hidden_states + gate_ff * ff_output[:, text_seq_length:]
+        encoder_hidden_states = encoder_hidden_states + enc_gate_ff * ff_output[:, :text_seq_length]
+
         return hidden_states, encoder_hidden_states
 
 
@@ -231,6 +234,7 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin):
         norm_eps: float = 1e-5,
         spatial_interpolation_scale: float = 1.875,
         temporal_interpolation_scale: float = 1.0,
+        use_rotary_positional_embeddings: bool = False,
     ):
         super().__init__()
         inner_dim = num_attention_heads * attention_head_dim
@@ -295,12 +299,113 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin):
     def _set_gradient_checkpointing(self, module, value=False):
         self.gradient_checkpointing = value
 
+    @property
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors
+    def attn_processors(self) -> Dict[str, AttentionProcessor]:
+        r"""
+        Returns:
+            `dict` of attention processors: A dictionary containing all attention processors used in the model with
+            indexed by its weight name.
+        """
+        # set recursively
+        processors = {}
+
+        def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]):
+            if hasattr(module, "get_processor"):
+                processors[f"{name}.processor"] = module.get_processor()
+
+            for sub_name, child in module.named_children():
+                fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
+
+            return processors
+
+        for name, module in self.named_children():
+            fn_recursive_add_processors(name, module, processors)
+
+        return processors
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor
+    def set_attn_processor(self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]):
+        r"""
+        Sets the attention processor to use to compute attention.
+
+        Parameters:
+            processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
+                The instantiated processor class or a dictionary of processor classes that will be set as the processor
+                for **all** `Attention` layers.
+
+                If `processor` is a dict, the key needs to define the path to the corresponding cross attention
+                processor. This is strongly recommended when setting trainable attention processors.
+
+        """
+        count = len(self.attn_processors.keys())
+
+        if isinstance(processor, dict) and len(processor) != count:
+            raise ValueError(
+                f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
+                f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
+            )
+
+        def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
+            if hasattr(module, "set_processor"):
+                if not isinstance(processor, dict):
+                    module.set_processor(processor)
+                else:
+                    module.set_processor(processor.pop(f"{name}.processor"))
+
+            for sub_name, child in module.named_children():
+                fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
+
+        for name, module in self.named_children():
+            fn_recursive_attn_processor(name, module, processor)
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections with FusedAttnProcessor2_0->FusedCogVideoXAttnProcessor2_0
+    def fuse_qkv_projections(self):
+        """
+        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query, key, value)
+        are fused. For cross-attention modules, key and value projection matrices are fused.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+        """
+        self.original_attn_processors = None
+
+        for _, attn_processor in self.attn_processors.items():
+            if "Added" in str(attn_processor.__class__.__name__):
+                raise ValueError("`fuse_qkv_projections()` is not supported for models having added KV projections.")
+
+        self.original_attn_processors = self.attn_processors
+
+        for module in self.modules():
+            if isinstance(module, Attention):
+                module.fuse_projections(fuse=True)
+
+        self.set_attn_processor(FusedCogVideoXAttnProcessor2_0())
+
+    # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.unfuse_qkv_projections
+    def unfuse_qkv_projections(self):
+        """Disables the fused QKV projection if enabled.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        """
+        if self.original_attn_processors is not None:
+            self.set_attn_processor(self.original_attn_processors)
+
     def forward(
         self,
         hidden_states: torch.Tensor,
         encoder_hidden_states: torch.Tensor,
         timestep: Union[int, float, torch.LongTensor],
         timestep_cond: Optional[torch.Tensor] = None,
+        image_rotary_emb: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
         return_dict: bool = True,
     ):
         batch_size, num_frames, channels, height, width = hidden_states.shape
@@ -319,14 +424,16 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin):
         hidden_states = self.patch_embed(encoder_hidden_states, hidden_states)
 
         # 3. Position embedding
-        seq_length = height * width * num_frames // (self.config.patch_size**2)
+        text_seq_length = encoder_hidden_states.shape[1]
+        if not self.config.use_rotary_positional_embeddings:
+            seq_length = height * width * num_frames // (self.config.patch_size**2)
 
-        pos_embeds = self.pos_embedding[:, : self.config.max_text_seq_length + seq_length]
-        hidden_states = hidden_states + pos_embeds
-        hidden_states = self.embedding_dropout(hidden_states)
+            pos_embeds = self.pos_embedding[:, : text_seq_length + seq_length]
+            hidden_states = hidden_states + pos_embeds
+            hidden_states = self.embedding_dropout(hidden_states)
 
-        encoder_hidden_states = hidden_states[:, : self.config.max_text_seq_length]
-        hidden_states = hidden_states[:, self.config.max_text_seq_length :]
+        encoder_hidden_states = hidden_states[:, :text_seq_length]
+        hidden_states = hidden_states[:, text_seq_length:]
 
         # 4. Transformer blocks
         for i, block in enumerate(self.transformer_blocks):
@@ -344,6 +451,7 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin):
                     hidden_states,
                     encoder_hidden_states,
                     emb,
+                    image_rotary_emb,
                     **ckpt_kwargs,
                 )
             else:
@@ -351,9 +459,17 @@ class CogVideoXTransformer3DModel(ModelMixin, ConfigMixin):
                     hidden_states=hidden_states,
                     encoder_hidden_states=encoder_hidden_states,
                     temb=emb,
+                    image_rotary_emb=image_rotary_emb,
                 )
 
-        hidden_states = self.norm_final(hidden_states)
+        if not self.config.use_rotary_positional_embeddings:
+            # CogVideoX-2B
+            hidden_states = self.norm_final(hidden_states)
+        else:
+            # CogVideoX-5B
+            hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
+            hidden_states = self.norm_final(hidden_states)
+            hidden_states = hidden_states[:, text_seq_length:]
 
         # 5. Final block
         hidden_states = self.norm_out(hidden_states, temb=emb)

src/diffusers/models/transformers/dit_transformer_2d.py

@@ -65,7 +65,6 @@ class DiTTransformer2DModel(ModelMixin, ConfigMixin):
     """
 
     _supports_gradient_checkpointing = True
-    _always_upcast_modules = ["PatchEmbed"]
 
     @register_to_config
     def __init__(

src/diffusers/models/transformers/hunyuan_transformer_2d.py

@@ -244,8 +244,6 @@ class HunyuanDiT2DModel(ModelMixin, ConfigMixin):
             Whether or not to use style condition and image meta size. True for version <=1.1, False for version >= 1.2
     """
 
-    _always_upcast_modules = ["HunyuanDiTAttentionPool"]
-
     @register_to_config
     def __init__(
         self,
@@ -486,9 +484,7 @@ class HunyuanDiT2DModel(ModelMixin, ConfigMixin):
         text_embedding_mask = torch.cat([text_embedding_mask, text_embedding_mask_t5], dim=-1)
         text_embedding_mask = text_embedding_mask.unsqueeze(2).bool()
 
-        encoder_hidden_states = torch.where(
-            text_embedding_mask, encoder_hidden_states, self.text_embedding_padding.to(encoder_hidden_states.dtype)
-        )
+        encoder_hidden_states = torch.where(text_embedding_mask, encoder_hidden_states, self.text_embedding_padding)
 
         skips = []
         for layer, block in enumerate(self.blocks):

src/diffusers/models/transformers/latte_transformer_3d.py

@@ -64,7 +64,6 @@ class LatteTransformer3DModel(ModelMixin, ConfigMixin):
         video_length (`int`, *optional*):
             The number of frames in the video-like data.
     """
-    _always_upcast_modules = ["PatchEmbed"]
 
     @register_to_config
     def __init__(
@@ -302,9 +301,7 @@ class LatteTransformer3DModel(ModelMixin, ConfigMixin):
                 hidden_states = hidden_states.reshape(-1, hidden_states.shape[-2], hidden_states.shape[-1])
 
         embedded_timestep = embedded_timestep.repeat_interleave(num_frame, dim=0).view(-1, embedded_timestep.shape[-1])
-        shift, scale = (self.scale_shift_table[None].to(embedded_timestep.dtype) + embedded_timestep[:, None]).chunk(
-            2, dim=1
-        )
+        shift, scale = (self.scale_shift_table[None] + embedded_timestep[:, None]).chunk(2, dim=1)
         hidden_states = self.norm_out(hidden_states)
         # Modulation
         hidden_states = hidden_states * (1 + scale) + shift

src/diffusers/models/transformers/pixart_transformer_2d.py

@@ -19,7 +19,7 @@ from torch import nn
 from ...configuration_utils import ConfigMixin, register_to_config
 from ...utils import is_torch_version, logging
 from ..attention import BasicTransformerBlock
-from ..attention_processor import Attention, AttentionProcessor, AttnProcessor, FusedAttnProcessor2_0
+from ..attention_processor import Attention, AttentionProcessor, FusedAttnProcessor2_0
 from ..embeddings import PatchEmbed, PixArtAlphaTextProjection
 from ..modeling_outputs import Transformer2DModelOutput
 from ..modeling_utils import ModelMixin
@@ -79,7 +79,6 @@ class PixArtTransformer2DModel(ModelMixin, ConfigMixin):
 
     _supports_gradient_checkpointing = True
     _no_split_modules = ["BasicTransformerBlock", "PatchEmbed"]
-    _always_upcast_modules = ["PatchEmbed"]
 
     @register_to_config
     def __init__(
@@ -248,14 +247,6 @@ class PixArtTransformer2DModel(ModelMixin, ConfigMixin):
         for name, module in self.named_children():
             fn_recursive_attn_processor(name, module, processor)
 
-    def set_default_attn_processor(self):
-        """
-        Disables custom attention processors and sets the default attention implementation.
-
-        Safe to just use `AttnProcessor()` as PixArt doesn't have any exotic attention processors in default model.
-        """
-        self.set_attn_processor(AttnProcessor())
-
     # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.fuse_qkv_projections
     def fuse_qkv_projections(self):
         """
@@ -423,8 +414,7 @@ class PixArtTransformer2DModel(ModelMixin, ConfigMixin):
 
         # 3. Output
         shift, scale = (
-            self.scale_shift_table[None].to(embedded_timestep.dtype)
-            + embedded_timestep[:, None].to(self.scale_shift_table.device)
+            self.scale_shift_table[None] + embedded_timestep[:, None].to(self.scale_shift_table.device)
         ).chunk(2, dim=1)
         hidden_states = self.norm_out(hidden_states)
         # Modulation

src/diffusers/models/transformers/prior_transformer.py

@@ -289,7 +289,7 @@ class PriorTransformer(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, Pef
 
         # timesteps does not contain any weights and will always return f32 tensors
         # but time_embedding might be fp16, so we need to cast here.
-        timesteps_projected = timesteps_projected.to(dtype=hidden_states.dtype)
+        timesteps_projected = timesteps_projected.to(dtype=self.dtype)
         time_embeddings = self.time_embedding(timesteps_projected)
 
         if self.embedding_proj_norm is not None:

src/diffusers/models/transformers/transformer_flux.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX Team. All rights reserved.
+# Copyright 2024 Black Forest Labs, The HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,7 +15,6 @@
 
 from typing import Any, Dict, List, Optional, Union
 
-import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -251,7 +250,6 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig
     """
 
     _supports_gradient_checkpointing = True
-    _no_split_modules = ["FluxTransformerBlock", "FluxSingleTransformerBlock"]
 
     @register_to_config
     def __init__(
@@ -323,8 +321,6 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig
         txt_ids: torch.Tensor = None,
         guidance: torch.Tensor = None,
         joint_attention_kwargs: Optional[Dict[str, Any]] = None,
-        controlnet_block_samples=None,
-        controlnet_single_block_samples=None,
         return_dict: bool = True,
     ) -> Union[torch.FloatTensor, Transformer2DModelOutput]:
         """
@@ -381,7 +377,6 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig
         )
         encoder_hidden_states = self.context_embedder(encoder_hidden_states)
 
-        txt_ids = txt_ids.expand(img_ids.size(0), -1, -1)
         ids = torch.cat((txt_ids, img_ids), dim=1)
         image_rotary_emb = self.pos_embed(ids)
 
@@ -415,12 +410,6 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig
                     image_rotary_emb=image_rotary_emb,
                 )
 
-            # controlnet residual
-            if controlnet_block_samples is not None:
-                interval_control = len(self.transformer_blocks) / len(controlnet_block_samples)
-                interval_control = int(np.ceil(interval_control))
-                hidden_states = hidden_states + controlnet_block_samples[index_block // interval_control]
-
         hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1)
 
         for index_block, block in enumerate(self.single_transformer_blocks):
@@ -451,15 +440,6 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig
                     image_rotary_emb=image_rotary_emb,
                 )
 
-            # controlnet residual
-            if controlnet_single_block_samples is not None:
-                interval_control = len(self.single_transformer_blocks) / len(controlnet_single_block_samples)
-                interval_control = int(np.ceil(interval_control))
-                hidden_states[:, encoder_hidden_states.shape[1] :, ...] = (
-                    hidden_states[:, encoder_hidden_states.shape[1] :, ...]
-                    + controlnet_single_block_samples[index_block // interval_control]
-                )
-
         hidden_states = hidden_states[:, encoder_hidden_states.shape[1] :, ...]
 
         hidden_states = self.norm_out(hidden_states, temb)

src/diffusers/models/transformers/transformer_sd3.py

@@ -54,7 +54,6 @@ class SD3Transformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrigi
     """
 
     _supports_gradient_checkpointing = True
-    _always_upcast_modules = ["PatchEmbed"]
 
     @register_to_config
     def __init__(

src/diffusers/models/unets/unet_2d.py

@@ -283,7 +283,7 @@ class UNet2DModel(ModelMixin, ConfigMixin):
         # timesteps does not contain any weights and will always return f32 tensors
         # but time_embedding might actually be running in fp16. so we need to cast here.
         # there might be better ways to encapsulate this.
-        t_emb = t_emb.to(dtype=sample.dtype)
+        t_emb = t_emb.to(dtype=self.dtype)
         emb = self.time_embedding(t_emb)
 
         if self.class_embedding is not None:

src/diffusers/models/unets/unet_3d_condition.py

@@ -641,7 +641,7 @@ class UNet3DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin)
         # timesteps does not contain any weights and will always return f32 tensors
         # but time_embedding might actually be running in fp16. so we need to cast here.
         # there might be better ways to encapsulate this.
-        t_emb = t_emb.to(dtype=sample.dtype)
+        t_emb = t_emb.to(dtype=self.dtype)
 
         emb = self.time_embedding(t_emb, timestep_cond)
         emb = emb.repeat_interleave(repeats=num_frames, dim=0)

src/diffusers/models/unets/unet_i2vgen_xl.py

@@ -590,7 +590,7 @@ class I2VGenXLUNet(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
         # timesteps does not contain any weights and will always return f32 tensors
         # but time_embedding might actually be running in fp16. so we need to cast here.
         # there might be better ways to encapsulate this.
-        t_emb = t_emb.to(dtype=sample.dtype)
+        t_emb = t_emb.to(dtype=self.dtype)
         t_emb = self.time_embedding(t_emb, timestep_cond)
 
         # 2. FPS

src/diffusers/models/unets/unet_motion_model.py

@@ -2152,7 +2152,7 @@ class UNetMotionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, Peft
         # timesteps does not contain any weights and will always return f32 tensors
         # but time_embedding might actually be running in fp16. so we need to cast here.
         # there might be better ways to encapsulate this.
-        t_emb = t_emb.to(dtype=sample.dtype)
+        t_emb = t_emb.to(dtype=self.dtype)
 
         emb = self.time_embedding(t_emb, timestep_cond)
         aug_emb = None

src/diffusers/pipelines/__init__.py

@@ -124,7 +124,7 @@ else:
         "AnimateDiffSparseControlNetPipeline",
         "AnimateDiffVideoToVideoPipeline",
     ]
-    _import_structure["flux"] = ["FluxPipeline", "FluxControlNetPipeline"]
+    _import_structure["flux"] = ["FluxPipeline"]
     _import_structure["audioldm"] = ["AudioLDMPipeline"]
     _import_structure["audioldm2"] = [
         "AudioLDM2Pipeline",
@@ -173,7 +173,6 @@ else:
     _import_structure["controlnet_sd3"].extend(
         [
             "StableDiffusion3ControlNetPipeline",
-            "StableDiffusion3ControlNetInpaintingPipeline",
         ]
     )
     _import_structure["deepfloyd_if"] = [
@@ -466,7 +465,9 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .controlnet_hunyuandit import (
             HunyuanDiTControlNetPipeline,
         )
-        from .controlnet_sd3 import StableDiffusion3ControlNetInpaintingPipeline, StableDiffusion3ControlNetPipeline
+        from .controlnet_sd3 import (
+            StableDiffusion3ControlNetPipeline,
+        )
         from .controlnet_xs import (
             StableDiffusionControlNetXSPipeline,
             StableDiffusionXLControlNetXSPipeline,
@@ -493,7 +494,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             VersatileDiffusionTextToImagePipeline,
             VQDiffusionPipeline,
         )
-        from .flux import FluxControlNetPipeline, FluxPipeline
+        from .flux import FluxPipeline
         from .hunyuandit import HunyuanDiTPipeline
         from .i2vgen_xl import I2VGenXLPipeline
         from .kandinsky import (

src/diffusers/pipelines/auto_pipeline.py

@@ -49,7 +49,6 @@ from .kandinsky2_2 import (
 )
 from .kandinsky3 import Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline
 from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
-from .lumina import LuminaText2ImgPipeline
 from .pag import (
     HunyuanDiTPAGPipeline,
     PixArtSigmaPAGPipeline,
@@ -107,7 +106,6 @@ AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
         ("pixart-sigma-pag", PixArtSigmaPAGPipeline),
         ("auraflow", AuraFlowPipeline),
         ("flux", FluxPipeline),
-        ("lumina", LuminaText2ImgPipeline),
     ]
 )
 

src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py

@@ -23,6 +23,7 @@ from transformers import T5EncoderModel, T5Tokenizer
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
+from ...models.embeddings import get_3d_rotary_pos_embed
 from ...pipelines.pipeline_utils import DiffusionPipeline
 from ...schedulers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler
 from ...utils import BaseOutput, logging, replace_example_docstring
@@ -40,6 +41,7 @@ EXAMPLE_DOC_STRING = """
         >>> from diffusers import CogVideoXPipeline
         >>> from diffusers.utils import export_to_video
 
+        >>> # Models: "THUDM/CogVideoX-2b" or "THUDM/CogVideoX-5b"
         >>> pipe = CogVideoXPipeline.from_pretrained("THUDM/CogVideoX-2b", torch_dtype=torch.float16).to("cuda")
         >>> prompt = (
         ...     "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. "
@@ -55,6 +57,25 @@ EXAMPLE_DOC_STRING = """
 """
 
 
+# Similar to diffusers.pipelines.hunyuandit.pipeline_hunyuandit.get_resize_crop_region_for_grid
+def get_resize_crop_region_for_grid(src, tgt_width, tgt_height):
+    tw = tgt_width
+    th = tgt_height
+    h, w = src
+    r = h / w
+    if r > (th / tw):
+        resize_height = th
+        resize_width = int(round(th / h * w))
+    else:
+        resize_width = tw
+        resize_height = int(round(tw / w * h))
+
+    crop_top = int(round((th - resize_height) / 2.0))
+    crop_left = int(round((tw - resize_width) / 2.0))
+
+    return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
 def retrieve_timesteps(
     scheduler,
@@ -409,6 +430,46 @@ class CogVideoXPipeline(DiffusionPipeline):
                     f" {negative_prompt_embeds.shape}."
                 )
 
+    def fuse_qkv_projections(self) -> None:
+        r"""Enables fused QKV projections."""
+        self.fusing_transformer = True
+        self.transformer.fuse_qkv_projections()
+
+    def unfuse_qkv_projections(self) -> None:
+        r"""Disable QKV projection fusion if enabled."""
+        if not self.fusing_transformer:
+            logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.")
+        else:
+            self.transformer.unfuse_qkv_projections()
+            self.fusing_transformer = False
+
+    def _prepare_rotary_positional_embeddings(
+        self,
+        height: int,
+        width: int,
+        num_frames: int,
+        device: torch.device,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+        grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+        base_size_width = 720 // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+        base_size_height = 480 // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+
+        grid_crops_coords = get_resize_crop_region_for_grid(
+            (grid_height, grid_width), base_size_width, base_size_height
+        )
+        freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+            embed_dim=self.transformer.config.attention_head_dim,
+            crops_coords=grid_crops_coords,
+            grid_size=(grid_height, grid_width),
+            temporal_size=num_frames,
+            use_real=True,
+        )
+
+        freqs_cos = freqs_cos.to(device=device)
+        freqs_sin = freqs_sin.to(device=device)
+        return freqs_cos, freqs_sin
+
     @property
     def guidance_scale(self):
         return self._guidance_scale
@@ -599,7 +660,14 @@ class CogVideoXPipeline(DiffusionPipeline):
         # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
-        # 7. Denoising loop
+        # 7. Create rotary embeds if required
+        image_rotary_emb = (
+            self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device)
+            if self.transformer.config.use_rotary_positional_embeddings
+            else None
+        )
+
+        # 8. Denoising loop
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
 
         with self.progress_bar(total=num_inference_steps) as progress_bar:
@@ -620,6 +688,7 @@ class CogVideoXPipeline(DiffusionPipeline):
                     hidden_states=latent_model_input,
                     encoder_hidden_states=prompt_embeds,
                     timestep=timestep,
+                    image_rotary_emb=image_rotary_emb,
                     return_dict=False,
                 )[0]
                 noise_pred = noise_pred.float()

src/diffusers/pipelines/controlnet_sd3/__init__.py

@@ -23,9 +23,6 @@ except OptionalDependencyNotAvailable:
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
     _import_structure["pipeline_stable_diffusion_3_controlnet"] = ["StableDiffusion3ControlNetPipeline"]
-    _import_structure["pipeline_stable_diffusion_3_controlnet_inpainting"] = [
-        "StableDiffusion3ControlNetInpaintingPipeline"
-    ]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
@@ -36,7 +33,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from ...utils.dummy_torch_and_transformers_objects import *
     else:
         from .pipeline_stable_diffusion_3_controlnet import StableDiffusion3ControlNetPipeline
-        from .pipeline_stable_diffusion_3_controlnet_inpainting import StableDiffusion3ControlNetInpaintingPipeline
 
     try:
         if not (is_transformers_available() and is_flax_available()):

src/diffusers/pipelines/flux/__init__.py

@@ -23,7 +23,6 @@ except OptionalDependencyNotAvailable:
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
     _import_structure["pipeline_flux"] = ["FluxPipeline"]
-    _import_structure["pipeline_flux_controlnet"] = ["FluxControlNetPipeline"]
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
         if not (is_transformers_available() and is_torch_available()):
@@ -32,7 +31,6 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
     else:
         from .pipeline_flux import FluxPipeline
-        from .pipeline_flux_controlnet import FluxControlNetPipeline
 else:
     import sys
 

src/diffusers/pipelines/flux/pipeline_flux.py

@@ -280,7 +280,7 @@ class FluxPipeline(DiffusionPipeline, FluxLoraLoaderMixin):
         prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
 
         # duplicate text embeddings for each generation per prompt, using mps friendly method
-        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt)
         prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1)
 
         return prompt_embeds
@@ -677,13 +677,6 @@ class FluxPipeline(DiffusionPipeline, FluxLoraLoaderMixin):
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
-        # handle guidance
-        if self.transformer.config.guidance_embeds:
-            guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
-            guidance = guidance.expand(latents.shape[0])
-        else:
-            guidance = None
-
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for i, t in enumerate(timesteps):
@@ -693,6 +686,13 @@ class FluxPipeline(DiffusionPipeline, FluxLoraLoaderMixin):
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                 timestep = t.expand(latents.shape[0]).to(latents.dtype)
 
+                # handle guidance
+                if self.transformer.config.guidance_embeds:
+                    guidance = torch.tensor([guidance_scale], device=device)
+                    guidance = guidance.expand(latents.shape[0])
+                else:
+                    guidance = None
+
                 noise_pred = self.transformer(
                     hidden_states=latents,
                     # YiYi notes: divide it by 1000 for now because we scale it by 1000 in the transforme rmodel (we should not keep it but I want to keep the inputs same for the model for testing)

src/diffusers/pipelines/flux/pipeline_flux_controlnet.py

@@ -1,861 +0,0 @@
-# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import inspect
-from typing import Any, Callable, Dict, List, Optional, Union
-
-import numpy as np
-import torch
-from transformers import (
-    CLIPTextModel,
-    CLIPTokenizer,
-    T5EncoderModel,
-    T5TokenizerFast,
-)
-
-from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FluxLoraLoaderMixin
-from ...models.autoencoders import AutoencoderKL
-from ...models.controlnet_flux import FluxControlNetModel
-from ...models.transformers import FluxTransformer2DModel
-from ...schedulers import FlowMatchEulerDiscreteScheduler
-from ...utils import (
-    USE_PEFT_BACKEND,
-    is_torch_xla_available,
-    logging,
-    replace_example_docstring,
-    scale_lora_layers,
-    unscale_lora_layers,
-)
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline
-from .pipeline_output import FluxPipelineOutput
-
-
-if is_torch_xla_available():
-    import torch_xla.core.xla_model as xm
-
-    XLA_AVAILABLE = True
-else:
-    XLA_AVAILABLE = False
-
-
-logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
-
-EXAMPLE_DOC_STRING = """
-    Examples:
-        ```py
-        >>> import torch
-        >>> from diffusers.utils import load_image
-        >>> from diffusers import FluxControlNetPipeline
-        >>> from diffusers import FluxControlNetModel
-
-        >>> controlnet_model = "InstantX/FLUX.1-dev-controlnet-canny-alpha"
-        >>> controlnet = FluxControlNetModel.from_pretrained(controlnet_model, torch_dtype=torch.bfloat16)
-        >>> pipe = FluxControlNetPipeline.from_pretrained(
-        ...     base_model, controlnet=controlnet, torch_dtype=torch.bfloat16
-        ... )
-        >>> pipe.to("cuda")
-        >>> control_image = load_image("https://huggingface.co/InstantX/SD3-Controlnet-Canny/resolve/main/canny.jpg")
-        >>> prompt = "A girl in city, 25 years old, cool, futuristic"
-        >>> image = pipe(
-        ...     prompt,
-        ...     control_image=control_image,
-        ...     controlnet_conditioning_scale=0.6,
-        ...     num_inference_steps=28,
-        ...     guidance_scale=3.5,
-        ... ).images[0]
-        >>> image.save("flux.png")
-        ```
-"""
-
-
-# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
-def calculate_shift(
-    image_seq_len,
-    base_seq_len: int = 256,
-    max_seq_len: int = 4096,
-    base_shift: float = 0.5,
-    max_shift: float = 1.16,
-):
-    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
-    b = base_shift - m * base_seq_len
-    mu = image_seq_len * m + b
-    return mu
-
-
-# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
-def retrieve_timesteps(
-    scheduler,
-    num_inference_steps: Optional[int] = None,
-    device: Optional[Union[str, torch.device]] = None,
-    timesteps: Optional[List[int]] = None,
-    sigmas: Optional[List[float]] = None,
-    **kwargs,
-):
-    """
-    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
-    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
-
-    Args:
-        scheduler (`SchedulerMixin`):
-            The scheduler to get timesteps from.
-        num_inference_steps (`int`):
-            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
-            must be `None`.
-        device (`str` or `torch.device`, *optional*):
-            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
-        timesteps (`List[int]`, *optional*):
-            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
-            `num_inference_steps` and `sigmas` must be `None`.
-        sigmas (`List[float]`, *optional*):
-            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
-            `num_inference_steps` and `timesteps` must be `None`.
-
-    Returns:
-        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
-        second element is the number of inference steps.
-    """
-    if timesteps is not None and sigmas is not None:
-        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
-    if timesteps is not None:
-        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
-        if not accepts_timesteps:
-            raise ValueError(
-                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
-                f" timestep schedules. Please check whether you are using the correct scheduler."
-            )
-        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
-        timesteps = scheduler.timesteps
-        num_inference_steps = len(timesteps)
-    elif sigmas is not None:
-        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
-        if not accept_sigmas:
-            raise ValueError(
-                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
-                f" sigmas schedules. Please check whether you are using the correct scheduler."
-            )
-        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
-        timesteps = scheduler.timesteps
-        num_inference_steps = len(timesteps)
-    else:
-        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
-        timesteps = scheduler.timesteps
-    return timesteps, num_inference_steps
-
-
-class FluxControlNetPipeline(DiffusionPipeline, FluxLoraLoaderMixin):
-    r"""
-    The Flux pipeline for text-to-image generation.
-
-    Reference: https://blackforestlabs.ai/announcing-black-forest-labs/
-
-    Args:
-        transformer ([`FluxTransformer2DModel`]):
-            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
-        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
-            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
-        vae ([`AutoencoderKL`]):
-            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
-        text_encoder ([`CLIPTextModel`]):
-            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
-            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
-        text_encoder_2 ([`T5EncoderModel`]):
-            [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically
-            the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant.
-        tokenizer (`CLIPTokenizer`):
-            Tokenizer of class
-            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
-        tokenizer_2 (`T5TokenizerFast`):
-            Second Tokenizer of class
-            [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast).
-    """
-
-    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
-    _optional_components = []
-    _callback_tensor_inputs = ["latents", "prompt_embeds"]
-
-    def __init__(
-        self,
-        scheduler: FlowMatchEulerDiscreteScheduler,
-        vae: AutoencoderKL,
-        text_encoder: CLIPTextModel,
-        tokenizer: CLIPTokenizer,
-        text_encoder_2: T5EncoderModel,
-        tokenizer_2: T5TokenizerFast,
-        transformer: FluxTransformer2DModel,
-        controlnet: FluxControlNetModel,
-    ):
-        super().__init__()
-
-        self.register_modules(
-            vae=vae,
-            text_encoder=text_encoder,
-            text_encoder_2=text_encoder_2,
-            tokenizer=tokenizer,
-            tokenizer_2=tokenizer_2,
-            transformer=transformer,
-            scheduler=scheduler,
-            controlnet=controlnet,
-        )
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels)) if hasattr(self, "vae") and self.vae is not None else 16
-        )
-        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
-        self.tokenizer_max_length = (
-            self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
-        )
-        self.default_sample_size = 64
-
-    def _get_t5_prompt_embeds(
-        self,
-        prompt: Union[str, List[str]] = None,
-        num_images_per_prompt: int = 1,
-        max_sequence_length: int = 512,
-        device: Optional[torch.device] = None,
-        dtype: Optional[torch.dtype] = None,
-    ):
-        device = device or self._execution_device
-        dtype = dtype or self.text_encoder.dtype
-
-        prompt = [prompt] if isinstance(prompt, str) else prompt
-        batch_size = len(prompt)
-
-        text_inputs = self.tokenizer_2(
-            prompt,
-            padding="max_length",
-            max_length=max_sequence_length,
-            truncation=True,
-            return_length=False,
-            return_overflowing_tokens=False,
-            return_tensors="pt",
-        )
-        text_input_ids = text_inputs.input_ids
-        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
-
-        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
-            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
-            logger.warning(
-                "The following part of your input was truncated because `max_sequence_length` is set to "
-                f" {max_sequence_length} tokens: {removed_text}"
-            )
-
-        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False)[0]
-
-        dtype = self.text_encoder_2.dtype
-        prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
-
-        _, seq_len, _ = prompt_embeds.shape
-
-        # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
-        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
-        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
-
-        return prompt_embeds
-
-    def _get_clip_prompt_embeds(
-        self,
-        prompt: Union[str, List[str]],
-        num_images_per_prompt: int = 1,
-        device: Optional[torch.device] = None,
-    ):
-        device = device or self._execution_device
-
-        prompt = [prompt] if isinstance(prompt, str) else prompt
-        batch_size = len(prompt)
-
-        text_inputs = self.tokenizer(
-            prompt,
-            padding="max_length",
-            max_length=self.tokenizer_max_length,
-            truncation=True,
-            return_overflowing_tokens=False,
-            return_length=False,
-            return_tensors="pt",
-        )
-
-        text_input_ids = text_inputs.input_ids
-        untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
-        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
-            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
-            logger.warning(
-                "The following part of your input was truncated because CLIP can only handle sequences up to"
-                f" {self.tokenizer_max_length} tokens: {removed_text}"
-            )
-        prompt_embeds = self.text_encoder(text_input_ids.to(device), output_hidden_states=False)
-
-        # Use pooled output of CLIPTextModel
-        prompt_embeds = prompt_embeds.pooler_output
-        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
-
-        # duplicate text embeddings for each generation per prompt, using mps friendly method
-        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
-        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1)
-
-        return prompt_embeds
-
-    def encode_prompt(
-        self,
-        prompt: Union[str, List[str]],
-        prompt_2: Union[str, List[str]],
-        device: Optional[torch.device] = None,
-        num_images_per_prompt: int = 1,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
-        max_sequence_length: int = 512,
-        lora_scale: Optional[float] = None,
-    ):
-        r"""
-
-        Args:
-            prompt (`str` or `List[str]`, *optional*):
-                prompt to be encoded
-            prompt_2 (`str` or `List[str]`, *optional*):
-                The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
-                used in all text-encoders
-            device: (`torch.device`):
-                torch device
-            num_images_per_prompt (`int`):
-                number of images that should be generated per prompt
-            prompt_embeds (`torch.FloatTensor`, *optional*):
-                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
-                provided, text embeddings will be generated from `prompt` input argument.
-            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
-                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
-                If not provided, pooled text embeddings will be generated from `prompt` input argument.
-            clip_skip (`int`, *optional*):
-                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
-                the output of the pre-final layer will be used for computing the prompt embeddings.
-            lora_scale (`float`, *optional*):
-                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
-        """
-        device = device or self._execution_device
-
-        # set lora scale so that monkey patched LoRA
-        # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, FluxLoraLoaderMixin):
-            self._lora_scale = lora_scale
-
-            # dynamically adjust the LoRA scale
-            if self.text_encoder is not None and USE_PEFT_BACKEND:
-                scale_lora_layers(self.text_encoder, lora_scale)
-            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
-                scale_lora_layers(self.text_encoder_2, lora_scale)
-
-        prompt = [prompt] if isinstance(prompt, str) else prompt
-        if prompt is not None:
-            batch_size = len(prompt)
-        else:
-            batch_size = prompt_embeds.shape[0]
-
-        if prompt_embeds is None:
-            prompt_2 = prompt_2 or prompt
-            prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
-
-            # We only use the pooled prompt output from the CLIPTextModel
-            pooled_prompt_embeds = self._get_clip_prompt_embeds(
-                prompt=prompt,
-                device=device,
-                num_images_per_prompt=num_images_per_prompt,
-            )
-            prompt_embeds = self._get_t5_prompt_embeds(
-                prompt=prompt_2,
-                num_images_per_prompt=num_images_per_prompt,
-                max_sequence_length=max_sequence_length,
-                device=device,
-            )
-
-        if self.text_encoder is not None:
-            if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
-                # Retrieve the original scale by scaling back the LoRA layers
-                unscale_lora_layers(self.text_encoder, lora_scale)
-
-        if self.text_encoder_2 is not None:
-            if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
-                # Retrieve the original scale by scaling back the LoRA layers
-                unscale_lora_layers(self.text_encoder_2, lora_scale)
-
-        dtype = self.text_encoder.dtype if self.text_encoder is not None else self.transformer.dtype
-        text_ids = torch.zeros(batch_size, prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
-        text_ids = text_ids.repeat(num_images_per_prompt, 1, 1)
-
-        return prompt_embeds, pooled_prompt_embeds, text_ids
-
-    def check_inputs(
-        self,
-        prompt,
-        prompt_2,
-        height,
-        width,
-        prompt_embeds=None,
-        pooled_prompt_embeds=None,
-        callback_on_step_end_tensor_inputs=None,
-        max_sequence_length=None,
-    ):
-        if height % 8 != 0 or width % 8 != 0:
-            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
-
-        if callback_on_step_end_tensor_inputs is not None and not all(
-            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
-        ):
-            raise ValueError(
-                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
-            )
-
-        if prompt is not None and prompt_embeds is not None:
-            raise ValueError(
-                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
-                " only forward one of the two."
-            )
-        elif prompt_2 is not None and prompt_embeds is not None:
-            raise ValueError(
-                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
-                " only forward one of the two."
-            )
-        elif prompt is None and prompt_embeds is None:
-            raise ValueError(
-                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
-            )
-        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
-            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
-        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
-            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
-
-        if prompt_embeds is not None and pooled_prompt_embeds is None:
-            raise ValueError(
-                "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
-            )
-
-        if max_sequence_length is not None and max_sequence_length > 512:
-            raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
-
-    @staticmethod
-    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._prepare_latent_image_ids
-    def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
-        latent_image_ids = torch.zeros(height // 2, width // 2, 3)
-        latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height // 2)[:, None]
-        latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width // 2)[None, :]
-
-        latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
-
-        latent_image_ids = latent_image_ids[None, :].repeat(batch_size, 1, 1, 1)
-        latent_image_ids = latent_image_ids.reshape(
-            batch_size, latent_image_id_height * latent_image_id_width, latent_image_id_channels
-        )
-
-        return latent_image_ids.to(device=device, dtype=dtype)
-
-    @staticmethod
-    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._pack_latents
-    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
-        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
-        latents = latents.permute(0, 2, 4, 1, 3, 5)
-        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)
-
-        return latents
-
-    @staticmethod
-    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._unpack_latents
-    def _unpack_latents(latents, height, width, vae_scale_factor):
-        batch_size, num_patches, channels = latents.shape
-
-        height = height // vae_scale_factor
-        width = width // vae_scale_factor
-
-        latents = latents.view(batch_size, height, width, channels // 4, 2, 2)
-        latents = latents.permute(0, 3, 1, 4, 2, 5)
-
-        latents = latents.reshape(batch_size, channels // (2 * 2), height * 2, width * 2)
-
-        return latents
-
-    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_latents
-    def prepare_latents(
-        self,
-        batch_size,
-        num_channels_latents,
-        height,
-        width,
-        dtype,
-        device,
-        generator,
-        latents=None,
-    ):
-        height = 2 * (int(height) // self.vae_scale_factor)
-        width = 2 * (int(width) // self.vae_scale_factor)
-
-        shape = (batch_size, num_channels_latents, height, width)
-
-        if latents is not None:
-            latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
-            return latents.to(device=device, dtype=dtype), latent_image_ids
-
-        if isinstance(generator, list) and len(generator) != batch_size:
-            raise ValueError(
-                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
-                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
-            )
-
-        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
-        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
-
-        latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
-
-        return latents, latent_image_ids
-
-    # Copied from diffusers.pipelines.controlnet_sd3.pipeline_stable_diffusion_3_controlnet.StableDiffusion3ControlNetPipeline.prepare_image
-    def prepare_image(
-        self,
-        image,
-        width,
-        height,
-        batch_size,
-        num_images_per_prompt,
-        device,
-        dtype,
-        do_classifier_free_guidance=False,
-        guess_mode=False,
-    ):
-        if isinstance(image, torch.Tensor):
-            pass
-        else:
-            image = self.image_processor.preprocess(image, height=height, width=width)
-
-        image_batch_size = image.shape[0]
-
-        if image_batch_size == 1:
-            repeat_by = batch_size
-        else:
-            # image batch size is the same as prompt batch size
-            repeat_by = num_images_per_prompt
-
-        image = image.repeat_interleave(repeat_by, dim=0)
-
-        image = image.to(device=device, dtype=dtype)
-
-        if do_classifier_free_guidance and not guess_mode:
-            image = torch.cat([image] * 2)
-
-        return image
-
-    @property
-    def guidance_scale(self):
-        return self._guidance_scale
-
-    @property
-    def joint_attention_kwargs(self):
-        return self._joint_attention_kwargs
-
-    @property
-    def num_timesteps(self):
-        return self._num_timesteps
-
-    @property
-    def interrupt(self):
-        return self._interrupt
-
-    @torch.no_grad()
-    @replace_example_docstring(EXAMPLE_DOC_STRING)
-    def __call__(
-        self,
-        prompt: Union[str, List[str]] = None,
-        prompt_2: Optional[Union[str, List[str]]] = None,
-        height: Optional[int] = None,
-        width: Optional[int] = None,
-        num_inference_steps: int = 28,
-        timesteps: List[int] = None,
-        guidance_scale: float = 7.0,
-        control_image: PipelineImageInput = None,
-        controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
-        num_images_per_prompt: Optional[int] = 1,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
-        output_type: Optional[str] = "pil",
-        return_dict: bool = True,
-        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
-        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
-        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
-        max_sequence_length: int = 512,
-    ):
-        r"""
-        Function invoked when calling the pipeline for generation.
-
-        Args:
-            prompt (`str` or `List[str]`, *optional*):
-                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
-                instead.
-            prompt_2 (`str` or `List[str]`, *optional*):
-                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
-                will be used instead
-            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The height in pixels of the generated image. This is set to 1024 by default for the best results.
-            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The width in pixels of the generated image. This is set to 1024 by default for the best results.
-            num_inference_steps (`int`, *optional*, defaults to 50):
-                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
-                expense of slower inference.
-            timesteps (`List[int]`, *optional*):
-                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
-                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
-                passed will be used. Must be in descending order.
-            guidance_scale (`float`, *optional*, defaults to 7.0):
-                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
-                `guidance_scale` is defined as `w` of equation 2. of [Imagen
-                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
-                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
-                usually at the expense of lower image quality.
-            num_images_per_prompt (`int`, *optional*, defaults to 1):
-                The number of images to generate per prompt.
-            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
-                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
-                to make generation deterministic.
-            latents (`torch.FloatTensor`, *optional*):
-                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
-                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
-                tensor will ge generated by sampling using the supplied random `generator`.
-            prompt_embeds (`torch.FloatTensor`, *optional*):
-                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
-                provided, text embeddings will be generated from `prompt` input argument.
-            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
-                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
-                If not provided, pooled text embeddings will be generated from `prompt` input argument.
-            output_type (`str`, *optional*, defaults to `"pil"`):
-                The output format of the generate image. Choose between
-                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
-            return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple.
-            joint_attention_kwargs (`dict`, *optional*):
-                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
-                `self.processor` in
-                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
-            callback_on_step_end (`Callable`, *optional*):
-                A function that calls at the end of each denoising steps during the inference. The function is called
-                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
-                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
-                `callback_on_step_end_tensor_inputs`.
-            callback_on_step_end_tensor_inputs (`List`, *optional*):
-                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
-                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
-                `._callback_tensor_inputs` attribute of your pipeline class.
-            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
-
-        Examples:
-
-        Returns:
-            [`~pipelines.flux.FluxPipelineOutput`] or `tuple`: [`~pipelines.flux.FluxPipelineOutput`] if `return_dict`
-            is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated
-            images.
-        """
-
-        height = height or self.default_sample_size * self.vae_scale_factor
-        width = width or self.default_sample_size * self.vae_scale_factor
-
-        # 1. Check inputs. Raise error if not correct
-        self.check_inputs(
-            prompt,
-            prompt_2,
-            height,
-            width,
-            prompt_embeds=prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
-            max_sequence_length=max_sequence_length,
-        )
-
-        self._guidance_scale = guidance_scale
-        self._joint_attention_kwargs = joint_attention_kwargs
-        self._interrupt = False
-
-        # 2. Define call parameters
-        if prompt is not None and isinstance(prompt, str):
-            batch_size = 1
-        elif prompt is not None and isinstance(prompt, list):
-            batch_size = len(prompt)
-        else:
-            batch_size = prompt_embeds.shape[0]
-
-        device = self._execution_device
-        dtype = self.transformer.dtype
-
-        lora_scale = (
-            self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
-        )
-        (
-            prompt_embeds,
-            pooled_prompt_embeds,
-            text_ids,
-        ) = self.encode_prompt(
-            prompt=prompt,
-            prompt_2=prompt_2,
-            prompt_embeds=prompt_embeds,
-            pooled_prompt_embeds=pooled_prompt_embeds,
-            device=device,
-            num_images_per_prompt=num_images_per_prompt,
-            max_sequence_length=max_sequence_length,
-            lora_scale=lora_scale,
-        )
-
-        # 3. Prepare control image
-        num_channels_latents = self.transformer.config.in_channels // 4
-        if isinstance(self.controlnet, FluxControlNetModel):
-            control_image = self.prepare_image(
-                image=control_image,
-                width=width,
-                height=height,
-                batch_size=batch_size * num_images_per_prompt,
-                num_images_per_prompt=num_images_per_prompt,
-                device=device,
-                dtype=dtype,
-            )
-            height, width = control_image.shape[-2:]
-
-            # vae encode
-            control_image = self.vae.encode(control_image).latent_dist.sample()
-            control_image = (control_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor
-
-            # pack
-            height_control_image, width_control_image = control_image.shape[2:]
-            control_image = self._pack_latents(
-                control_image,
-                batch_size * num_images_per_prompt,
-                num_channels_latents,
-                height_control_image,
-                width_control_image,
-            )
-
-        # 4. Prepare latent variables
-        num_channels_latents = self.transformer.config.in_channels // 4
-        latents, latent_image_ids = self.prepare_latents(
-            batch_size * num_images_per_prompt,
-            num_channels_latents,
-            height,
-            width,
-            prompt_embeds.dtype,
-            device,
-            generator,
-            latents,
-        )
-
-        # 5. Prepare timesteps
-        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
-        image_seq_len = latents.shape[1]
-        mu = calculate_shift(
-            image_seq_len,
-            self.scheduler.config.base_image_seq_len,
-            self.scheduler.config.max_image_seq_len,
-            self.scheduler.config.base_shift,
-            self.scheduler.config.max_shift,
-        )
-        timesteps, num_inference_steps = retrieve_timesteps(
-            self.scheduler,
-            num_inference_steps,
-            device,
-            timesteps,
-            sigmas,
-            mu=mu,
-        )
-
-        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
-        self._num_timesteps = len(timesteps)
-
-        # 6. Denoising loop
-        with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for i, t in enumerate(timesteps):
-                if self.interrupt:
-                    continue
-
-                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
-                timestep = t.expand(latents.shape[0]).to(latents.dtype)
-
-                # handle guidance
-                if self.transformer.config.guidance_embeds:
-                    guidance = torch.tensor([guidance_scale], device=device)
-                    guidance = guidance.expand(latents.shape[0])
-                else:
-                    guidance = None
-
-                # controlnet
-                controlnet_block_samples, controlnet_single_block_samples = self.controlnet(
-                    hidden_states=latents,
-                    controlnet_cond=control_image,
-                    conditioning_scale=controlnet_conditioning_scale,
-                    timestep=timestep / 1000,
-                    guidance=guidance,
-                    pooled_projections=pooled_prompt_embeds,
-                    encoder_hidden_states=prompt_embeds,
-                    txt_ids=text_ids,
-                    img_ids=latent_image_ids,
-                    joint_attention_kwargs=self.joint_attention_kwargs,
-                    return_dict=False,
-                )
-
-                noise_pred = self.transformer(
-                    hidden_states=latents,
-                    # YiYi notes: divide it by 1000 for now because we scale it by 1000 in the transforme rmodel (we should not keep it but I want to keep the inputs same for the model for testing)
-                    timestep=timestep / 1000,
-                    guidance=guidance,
-                    pooled_projections=pooled_prompt_embeds,
-                    encoder_hidden_states=prompt_embeds,
-                    controlnet_block_samples=controlnet_block_samples,
-                    controlnet_single_block_samples=controlnet_single_block_samples,
-                    txt_ids=text_ids,
-                    img_ids=latent_image_ids,
-                    joint_attention_kwargs=self.joint_attention_kwargs,
-                    return_dict=False,
-                )[0]
-
-                # compute the previous noisy sample x_t -> x_t-1
-                latents_dtype = latents.dtype
-                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
-
-                if latents.dtype != latents_dtype:
-                    if torch.backends.mps.is_available():
-                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
-                        latents = latents.to(latents_dtype)
-
-                if callback_on_step_end is not None:
-                    callback_kwargs = {}
-                    for k in callback_on_step_end_tensor_inputs:
-                        callback_kwargs[k] = locals()[k]
-                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
-
-                    latents = callback_outputs.pop("latents", latents)
-                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
-
-                # call the callback, if provided
-                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
-                    progress_bar.update()
-
-                if XLA_AVAILABLE:
-                    xm.mark_step()
-
-        if output_type == "latent":
-            image = latents
-
-        else:
-            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
-            latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
-
-            image = self.vae.decode(latents, return_dict=False)[0]
-            image = self.image_processor.postprocess(image, output_type=output_type)
-
-        # Offload all models
-        self.maybe_free_model_hooks()
-
-        if not return_dict:
-            return (image,)
-
-        return FluxPipelineOutput(images=image)