decorate UNetControlNetXSModelTests::test_forward_no_control with is_flaky

[PixArt] fix small nits in pixart sigma (#7767 )
fix small nits in pixart sigma
2026-01-01 09:21:06 +08:00 · 2024-04-25 07:03:42 +05:30 · 2024-04-25 06:37:35 +05:30 · 2024-04-23 22:33:08 -10:00 · 2024-04-24 12:28:11 +05:30 · 2024-04-24 12:27:45 +05:30
40 changed files with 5902 additions and 519 deletions
--- a/docs/source/en/_toctree.yml
+++ b/docs/source/en/_toctree.yml
@@ -23,156 +23,146 @@
    title: Accelerate inference of text-to-image diffusion models
  title: Tutorials
 - sections:
-  - sections:
-    - local: using-diffusers/loading
-      title: Load pipelines
-    - local: using-diffusers/custom_pipeline_overview
-      title: Load community pipelines and components
-    - local: using-diffusers/schedulers
-      title: Load schedulers and models
-    - local: using-diffusers/using_safetensors
-      title: Load safetensors
-    - local: using-diffusers/other-formats
-      title: Load different Stable Diffusion formats
-    - local: using-diffusers/loading_adapters
-      title: Load adapters
-    - local: using-diffusers/push_to_hub
-      title: Push files to the Hub
-    title: Loading & Hub
-  - sections:
-    - local: using-diffusers/pipeline_overview
-      title: Overview
-    - local: using-diffusers/unconditional_image_generation
-      title: Unconditional image generation
-    - local: using-diffusers/conditional_image_generation
-      title: Text-to-image
-    - local: using-diffusers/img2img
-      title: Image-to-image
-    - local: using-diffusers/inpaint
-      title: Inpainting
-    - local: using-diffusers/text-img2vid
-      title: Text or image-to-video
-    - local: using-diffusers/depth2img
-      title: Depth-to-image
-    title: Tasks
-  - sections:
-    - local: using-diffusers/textual_inversion_inference
-      title: Textual inversion
-    - local: using-diffusers/ip_adapter
-      title: IP-Adapter
-    - local: using-diffusers/merge_loras
-      title: Merge LoRAs
-    - local: training/distributed_inference
-      title: Distributed inference with multiple GPUs
-    - local: using-diffusers/reusing_seeds
-      title: Improve image quality with deterministic generation
-    - local: using-diffusers/control_brightness
-      title: Control image brightness
-    - local: using-diffusers/weighted_prompts
-      title: Prompt techniques
-    - local: using-diffusers/freeu
-      title: Improve generation quality with FreeU
-    title: Techniques
-  - sections:
-    - local: using-diffusers/pipeline_overview
-      title: Overview
-    - local: using-diffusers/sdxl
-      title: Stable Diffusion XL
-    - local: using-diffusers/sdxl_turbo
-      title: SDXL Turbo
-    - local: using-diffusers/kandinsky
-      title: Kandinsky
-    - local: using-diffusers/controlnet
-      title: ControlNet
-    - local: using-diffusers/t2i_adapter
-      title: T2I-Adapter
-    - local: using-diffusers/shap-e
-      title: Shap-E
-    - local: using-diffusers/diffedit
-      title: DiffEdit
-    - local: using-diffusers/distilled_sd
-      title: Distilled Stable Diffusion inference
-    - local: using-diffusers/callback
-      title: Pipeline callbacks
-    - local: using-diffusers/reproducibility
-      title: Create reproducible pipelines
-    - local: using-diffusers/custom_pipeline_examples
-      title: Community pipelines
-    - local: using-diffusers/contribute_pipeline
-      title: Contribute a community pipeline
-    - local: using-diffusers/inference_with_lcm_lora
-      title: Latent Consistency Model-LoRA
-    - local: using-diffusers/inference_with_lcm
-      title: Latent Consistency Model
-    - local: using-diffusers/inference_with_tcd_lora
-      title: Trajectory Consistency Distillation-LoRA
-    - local: using-diffusers/svd
-      title: Stable Video Diffusion
-    title: Specific pipeline examples
-  - sections:
-    - local: training/overview
-      title: Overview
-    - local: training/create_dataset
-      title: Create a dataset for training
-    - local: training/adapt_a_model
-      title: Adapt a model to a new task
-    - sections:
-      - local: training/unconditional_training
-        title: Unconditional image generation
-      - local: training/text2image
-        title: Text-to-image
-      - local: training/sdxl
-        title: Stable Diffusion XL
-      - local: training/kandinsky
-        title: Kandinsky 2.2
-      - local: training/wuerstchen
-        title: Wuerstchen
-      - local: training/controlnet
-        title: ControlNet
-      - local: training/t2i_adapters
-        title: T2I-Adapters
-      - local: training/instructpix2pix
-        title: InstructPix2Pix
-      title: Models
-    - sections:
-      - local: training/text_inversion
-        title: Textual Inversion
-      - local: training/dreambooth
-        title: DreamBooth
-      - local: training/lora
-        title: LoRA
-      - local: training/custom_diffusion
-        title: Custom Diffusion
-      - local: training/lcm_distill
-        title: Latent Consistency Distillation
-      - local: training/ddpo
-        title: Reinforcement learning training with DDPO
-      title: Methods
-    title: Training
-  - sections:
-    - local: using-diffusers/other-modalities
-      title: Other Modalities
-    title: Taking Diffusers Beyond Images
-  title: Using Diffusers
+  - local: using-diffusers/loading
+    title: Load pipelines
+  - local: using-diffusers/custom_pipeline_overview
+    title: Load community pipelines and components
+  - local: using-diffusers/schedulers
+    title: Load schedulers and models
+  - local: using-diffusers/using_safetensors
+    title: Load safetensors
+  - local: using-diffusers/other-formats
+    title: Load different Stable Diffusion formats
+  - local: using-diffusers/loading_adapters
+    title: Load adapters
+  - local: using-diffusers/push_to_hub
+    title: Push files to the Hub
+  title: Load pipelines and adapters
 - sections:
-  - local: optimization/opt_overview
+  - local: using-diffusers/unconditional_image_generation
+    title: Unconditional image generation
+  - local: using-diffusers/conditional_image_generation
+    title: Text-to-image
+  - local: using-diffusers/img2img
+    title: Image-to-image
+  - local: using-diffusers/inpaint
+    title: Inpainting
+  - local: using-diffusers/text-img2vid
+    title: Text or image-to-video
+  - local: using-diffusers/depth2img
+    title: Depth-to-image
+  title: Generative tasks
+- sections:
+  - local: using-diffusers/overview_techniques
    title: Overview
+  - local: training/distributed_inference
+    title: Distributed inference with multiple GPUs
+  - local: using-diffusers/merge_loras
+    title: Merge LoRAs
+  - local: using-diffusers/callback
+    title: Pipeline callbacks
+  - local: using-diffusers/reusing_seeds
+    title: Improve image quality with deterministic generation
+  - local: using-diffusers/control_brightness
+    title: Control image brightness
+  - local: using-diffusers/weighted_prompts
+    title: Prompt techniques
+  - local: using-diffusers/freeu
+    title: Improve generation quality with FreeU
+  title: Inference techniques
+- sections:
+  - local: using-diffusers/sdxl
+    title: Stable Diffusion XL
+  - local: using-diffusers/sdxl_turbo
+    title: SDXL Turbo
+  - local: using-diffusers/kandinsky
+    title: Kandinsky
+  - local: using-diffusers/ip_adapter
+    title: IP-Adapter
+  - local: using-diffusers/controlnet
+    title: ControlNet
+  - local: using-diffusers/t2i_adapter
+    title: T2I-Adapter
+  - local: using-diffusers/textual_inversion_inference
+    title: Textual inversion
+  - local: using-diffusers/shap-e
+    title: Shap-E
+  - local: using-diffusers/diffedit
+    title: DiffEdit
+  - local: using-diffusers/reproducibility
+    title: Create reproducible pipelines
+  - local: using-diffusers/custom_pipeline_examples
+    title: Community pipelines
+  - local: using-diffusers/contribute_pipeline
+    title: Contribute a community pipeline
+  - local: using-diffusers/inference_with_lcm_lora
+    title: Latent Consistency Model-LoRA
+  - local: using-diffusers/inference_with_lcm
+    title: Latent Consistency Model
+  - local: using-diffusers/inference_with_tcd_lora
+    title: Trajectory Consistency Distillation-LoRA
+  - local: using-diffusers/svd
+    title: Stable Video Diffusion
+  title: Specific pipeline examples
+- sections:
+  - local: training/overview
+    title: Overview
+  - local: training/create_dataset
+    title: Create a dataset for training
+  - local: training/adapt_a_model
+    title: Adapt a model to a new task
  - sections:
-    - local: optimization/fp16
-      title: Speed up inference
-    - local: optimization/memory
-      title: Reduce memory usage
-    - local: optimization/torch2.0
-      title: PyTorch 2.0
-    - local: optimization/xformers
-      title: xFormers
-    - local: optimization/tome
-      title: Token merging
-    - local: optimization/deepcache
-      title: DeepCache
-    - local: optimization/tgate
-      title: TGATE
-    title: General optimizations
+    - local: training/unconditional_training
+      title: Unconditional image generation
+    - local: training/text2image
+      title: Text-to-image
+    - local: training/sdxl
+      title: Stable Diffusion XL
+    - local: training/kandinsky
+      title: Kandinsky 2.2
+    - local: training/wuerstchen
+      title: Wuerstchen
+    - local: training/controlnet
+      title: ControlNet
+    - local: training/t2i_adapters
+      title: T2I-Adapters
+    - local: training/instructpix2pix
+      title: InstructPix2Pix
+    title: Models
+    isExpanded: false
+  - sections:
+    - local: training/text_inversion
+      title: Textual Inversion
+    - local: training/dreambooth
+      title: DreamBooth
+    - local: training/lora
+      title: LoRA
+    - local: training/custom_diffusion
+      title: Custom Diffusion
+    - local: training/lcm_distill
+      title: Latent Consistency Distillation
+    - local: training/ddpo
+      title: Reinforcement learning training with DDPO
+    title: Methods
+    isExpanded: false
+  title: Training
+- sections:
+  - local: optimization/fp16
+    title: Speed up inference
+  - local: using-diffusers/distilled_sd
+    title: Distilled Stable Diffusion inference
+  - local: optimization/memory
+    title: Reduce memory usage
+  - local: optimization/torch2.0
+    title: PyTorch 2.0
+  - local: optimization/xformers
+    title: xFormers
+  - local: optimization/tome
+    title: Token merging
+  - local: optimization/deepcache
+    title: DeepCache
+  - local: optimization/tgate
+    title: TGATE
  - sections:
    - local: using-diffusers/stable_diffusion_jax_how_to
      title: JAX/Flax
@@ -182,14 +172,14 @@
      title: OpenVINO
    - local: optimization/coreml
      title: Core ML
-    title: Optimized model types
+    title: Optimized model formats
  - sections:
    - local: optimization/mps
      title: Metal Performance Shaders (MPS)
    - local: optimization/habana
      title: Habana Gaudi
    title: Optimized hardware
-  title: Optimization
+  title: Accelerate inference and reduce memory
 - sections:
  - local: conceptual/philosophy
    title: Philosophy
@@ -211,6 +201,7 @@
    - local: api/outputs
      title: Outputs
    title: Main Classes
+    isExpanded: false
  - sections:
    - local: api/loaders/ip_adapter
      title: IP-Adapter
@@ -225,6 +216,7 @@
    - local: api/loaders/peft
      title: PEFT
    title: Loaders
+    isExpanded: false
  - sections:
    - local: api/models/overview
      title: Overview
@@ -259,6 +251,7 @@
    - local: api/models/controlnet
      title: ControlNet
    title: Models
+    isExpanded: false
  - sections:
    - local: api/pipelines/overview
      title: Overview
@@ -383,6 +376,7 @@
    - local: api/pipelines/wuerstchen
      title: Wuerstchen
    title: Pipelines
+    isExpanded: false
  - sections:
    - local: api/schedulers/overview
      title: Overview
@@ -443,6 +437,7 @@
    - local: api/schedulers/vq_diffusion
      title: VQDiffusionScheduler
    title: Schedulers
+    isExpanded: false
  - sections:
    - local: api/internal_classes_overview
      title: Overview
@@ -457,4 +452,5 @@
    - local: api/image_processor
      title: VAE Image Processor
    title: Internal classes
+    isExpanded: false
  title: API
--- a/docs/source/en/api/pipelines/auto_pipeline.md
+++ b/docs/source/en/api/pipelines/auto_pipeline.md
@@ -12,42 +12,10 @@ specific language governing permissions and limitations under the License.

 # AutoPipeline

-`AutoPipeline` is designed to:
-
-1. make it easy for you to load a checkpoint for a task without knowing the specific pipeline class to use
-2. use multiple pipelines in your workflow
-
-Based on the task, the `AutoPipeline` class automatically retrieves the relevant pipeline given the name or path to the pretrained weights with the `from_pretrained()` method.
-
-To seamlessly switch between tasks with the same checkpoint without reallocating additional memory, use the `from_pipe()` method to transfer the components from the original pipeline to the new one.
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True
-).to("cuda")
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-image = pipeline(prompt, num_inference_steps=25).images[0]
-```
-
-<Tip>
-
-Check out the [AutoPipeline](../../tutorials/autopipeline) tutorial to learn how to use this API!
-
-</Tip>
-
-`AutoPipeline` supports text-to-image, image-to-image, and inpainting for the following diffusion models:
-
- [Stable Diffusion](./stable_diffusion/overview)
- [ControlNet](./controlnet)
- [Stable Diffusion XL (SDXL)](./stable_diffusion/stable_diffusion_xl)
- [DeepFloyd IF](./deepfloyd_if)
- [Kandinsky 2.1](./kandinsky)
- [Kandinsky 2.2](./kandinsky_v22)
+The `AutoPipeline` is designed to make it easy to load a checkpoint for a task without needing to know the specific pipeline class. Based on the task, the `AutoPipeline` automatically retrieves the correct pipeline class from the checkpoint `model_index.json` file.

+> [!TIP]
+> Check out the [AutoPipeline](../../tutorials/autopipeline) tutorial to learn how to use this API!

 ## AutoPipelineForText2Image

--- a/docs/source/en/optimization/opt_overview.md
+++ b/docs/source/en/optimization/opt_overview.md
@@ -1,17 +0,0 @@
-<!--Copyright 2024 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. 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.
-->
-
-# Overview
-
-Generating high-quality outputs is computationally intensive, especially during each iterative step where you go from a noisy output to a less noisy output. One of 🤗 Diffuser's goals is to make this technology widely accessible to everyone, which includes enabling fast inference on consumer and specialized hardware.
-
-This section will cover tips and tricks - like half-precision weights and sliced attention - for optimizing inference speed and reducing memory-consumption. You'll also learn how to speed up your PyTorch code with [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) or [ONNX Runtime](https://onnxruntime.ai/docs/), and enable memory-efficient attention with [xFormers](https://facebookresearch.github.io/xformers/). There are also guides for running inference on specific hardware like Apple Silicon, and Intel or Habana processors.
--- a/docs/source/en/tutorials/autopipeline.md
+++ b/docs/source/en/tutorials/autopipeline.md
@@ -12,75 +12,74 @@ specific language governing permissions and limitations under the License.

 # AutoPipeline

-🤗 Diffusers is able to complete many different tasks, and you can often reuse the same pretrained weights for multiple tasks such as text-to-image, image-to-image, and inpainting. If you're new to the library and diffusion models though, it may be difficult to know which pipeline to use for a task. For example, if you're using the [runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5) checkpoint for text-to-image, you might not know that you could also use it for image-to-image and inpainting by loading the checkpoint with the [`StableDiffusionImg2ImgPipeline`] and [`StableDiffusionInpaintPipeline`] classes respectively.
+Diffusers provides many pipelines for basic tasks like generating images, videos, audio, and inpainting. On top of these, there are specialized pipelines for adapters and features like upscaling, super-resolution, and more. Different pipeline classes can even use the same checkpoint because they share the same pretrained model! With so many different pipelines, it can be overwhelming to know which pipeline class to use.

-The `AutoPipeline` class is designed to simplify the variety of pipelines in 🤗 Diffusers. It is a generic, *task-first* pipeline that lets you focus on the task. The `AutoPipeline` automatically detects the correct pipeline class to use, which makes it easier to load a checkpoint for a task without knowing the specific pipeline class name.
+The [AutoPipeline](../api/pipelines/auto_pipeline) class is designed to simplify the variety of pipelines in Diffusers. It is a generic *task-first* pipeline that lets you focus on a task ([`AutoPipelineForText2Image`], [`AutoPipelineForImage2Image`], and [`AutoPipelineForInpainting`]) without needing to know the specific pipeline class. The [AutoPipeline](../api/pipelines/auto_pipeline) automatically detects the correct pipeline class to use.

-<Tip>
+For example, let's use the [dreamlike-art/dreamlike-photoreal-2.0](https://hf.co/dreamlike-art/dreamlike-photoreal-2.0) checkpoint.

-Take a look at the [AutoPipeline](../api/pipelines/auto_pipeline) reference to see which tasks are supported. Currently, it supports text-to-image, image-to-image, and inpainting.
+Under the hood, [AutoPipeline](../api/pipelines/auto_pipeline):

-</Tip>
+1. Detects a `"stable-diffusion"` class from the [model_index.json](https://hf.co/dreamlike-art/dreamlike-photoreal-2.0/blob/main/model_index.json) file.
+2. Depending on the task you're interested in, it loads the [`StableDiffusionPipeline`], [`StableDiffusionImg2ImgPipeline`], or [`StableDiffusionInpaintPipeline`]. Any parameter (`strength`, `num_inference_steps`, etc.) you would pass to these specific pipelines can also be passed to the [AutoPipeline](../api/pipelines/auto_pipeline).

-This tutorial shows you how to use an `AutoPipeline` to automatically infer the pipeline class to load for a specific task, given the pretrained weights.
-
-## Choose an AutoPipeline for your task
-
-Start by picking a checkpoint. For example, if you're interested in text-to-image with the [runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5) checkpoint, use [`AutoPipelineForText2Image`]:
+<hfoptions id="autopipeline">
+<hfoption id="text-to-image">

 ```py
 from diffusers import AutoPipelineForText2Image
 import torch

-pipeline = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True
+pipe_txt2img = AutoPipelineForText2Image.from_pretrained(
+    "dreamlike-art/dreamlike-photoreal-2.0", torch_dtype=torch.float16, use_safetensors=True
 ).to("cuda")
-prompt = "peasant and dragon combat, wood cutting style, viking era, bevel with rune"

-image = pipeline(prompt, num_inference_steps=25).images[0]
+prompt = "cinematic photo of Godzilla eating sushi with a cat in a izakaya, 35mm photograph, film, professional, 4k, highly detailed"
+generator = torch.Generator(device="cpu").manual_seed(37)
+image = pipe_txt2img(prompt, generator=generator).images[0]
 image
 ```

 <div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-text2img.png" alt="generated image of peasant fighting dragon in wood cutting style"/>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-text2img.png"/>
 </div>

-Under the hood, [`AutoPipelineForText2Image`]:
-
-1. automatically detects a `"stable-diffusion"` class from the [`model_index.json`](https://huggingface.co/runwayml/stable-diffusion-v1-5/blob/main/model_index.json) file
-2. loads the corresponding text-to-image [`StableDiffusionPipeline`] based on the `"stable-diffusion"` class name
-
-Likewise, for image-to-image, [`AutoPipelineForImage2Image`] detects a `"stable-diffusion"` checkpoint from the `model_index.json` file and it'll load the corresponding [`StableDiffusionImg2ImgPipeline`] behind the scenes. You can also pass any additional arguments specific to the pipeline class such as `strength`, which determines the amount of noise or variation added to an input image:
+</hfoption>
+<hfoption id="image-to-image">

 ```py
 from diffusers import AutoPipelineForImage2Image
+from diffusers.utils import load_image
 import torch
-import requests
-from PIL import Image
-from io import BytesIO

-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
+pipe_img2img = AutoPipelineForImage2Image.from_pretrained(
+    "dreamlike-art/dreamlike-photoreal-2.0", torch_dtype=torch.float16, use_safetensors=True
 ).to("cuda")
-prompt = "a portrait of a dog wearing a pearl earring"

-url = "https://upload.wikimedia.org/wikipedia/commons/thumb/0/0f/1665_Girl_with_a_Pearl_Earring.jpg/800px-1665_Girl_with_a_Pearl_Earring.jpg"
+init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-text2img.png")

-response = requests.get(url)
-image = Image.open(BytesIO(response.content)).convert("RGB")
-image.thumbnail((768, 768))
-
-image = pipeline(prompt, image, num_inference_steps=200, strength=0.75, guidance_scale=10.5).images[0]
+prompt = "cinematic photo of Godzilla eating burgers with a cat in a fast food restaurant, 35mm photograph, film, professional, 4k, highly detailed"
+generator = torch.Generator(device="cpu").manual_seed(53)
+image = pipe_img2img(prompt, image=init_image, generator=generator).images[0]
 image
 ```

+Notice how the [dreamlike-art/dreamlike-photoreal-2.0](https://hf.co/dreamlike-art/dreamlike-photoreal-2.0) checkpoint is used for both text-to-image and image-to-image tasks? To save memory and avoid loading the checkpoint twice, use the [`~DiffusionPipeline.from_pipe`] method.
+
+```py
+pipe_img2img = AutoPipelineForImage2Image.from_pipe(pipe_txt2img).to("cuda")
+image = pipeline(prompt, image=init_image, generator=generator).images[0]
+image
+```
+
+You can learn more about the [`~DiffusionPipeline.from_pipe`] method in the [Reuse a pipeline](../using-diffusers/loading#reuse-a-pipeline) guide.
+
 <div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-img2img.png" alt="generated image of a vermeer portrait of a dog wearing a pearl earring"/>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-img2img.png"/>
 </div>

-And if you want to do inpainting, then [`AutoPipelineForInpainting`] loads the underlying [`StableDiffusionInpaintPipeline`] class in the same way:
+</hfoption>
+<hfoption id="inpainting">

 ```py
 from diffusers import AutoPipelineForInpainting
@@ -91,22 +90,27 @@ pipeline = AutoPipelineForInpainting.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, use_safetensors=True
 ).to("cuda")

-img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
-mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-img2img.png")
+mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-mask.png")

-init_image = load_image(img_url).convert("RGB")
-mask_image = load_image(mask_url).convert("RGB")
-
-prompt = "A majestic tiger sitting on a bench"
-image = pipeline(prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80).images[0]
+prompt = "cinematic photo of a owl, 35mm photograph, film, professional, 4k, highly detailed"
+generator = torch.Generator(device="cpu").manual_seed(38)
+image = pipeline(prompt, image=init_image, mask_image=mask_image, generator=generator, strength=0.4).images[0]
 image
 ```

 <div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-inpaint.png" alt="generated image of a tiger sitting on a bench"/>
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/autopipeline-inpaint.png"/>
 </div>

-If you try to load an unsupported checkpoint, it'll throw an error:
+</hfoption>
+</hfoptions>
+
+## Unsupported checkpoints
+
+The [AutoPipeline](../api/pipelines/auto_pipeline) supports [Stable Diffusion](../api/pipelines/stable_diffusion/overview), [Stable Diffusion XL](../api/pipelines/stable_diffusion/stable_diffusion_xl), [ControlNet](../api/pipelines/controlnet), [Kandinsky 2.1](../api/pipelines/kandinsky.md), [Kandinsky 2.2](../api/pipelines/kandinsky_v22), and [DeepFloyd IF](../api/pipelines/deepfloyd_if) checkpoints.
+
+If you try to load an unsupported checkpoint, you'll get an error.

 ```py
 from diffusers import AutoPipelineForImage2Image
@@ -117,54 +121,3 @@ pipeline = AutoPipelineForImage2Image.from_pretrained(
 )
 "ValueError: AutoPipeline can't find a pipeline linked to ShapEImg2ImgPipeline for None"
 ```
-
-## Use multiple pipelines
-
-For some workflows or if you're loading many pipelines, it is more memory-efficient to reuse the same components from a checkpoint instead of reloading them which would unnecessarily consume additional memory. For example, if you're using a checkpoint for text-to-image and you want to use it again for image-to-image, use the [`~AutoPipelineForImage2Image.from_pipe`] method. This method creates a new pipeline from the components of a previously loaded pipeline at no additional memory cost.
-
-The [`~AutoPipelineForImage2Image.from_pipe`] method detects the original pipeline class and maps it to the new pipeline class corresponding to the task you want to do. For example, if you load a `"stable-diffusion"` class pipeline for text-to-image:
-
-```py
-from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
-import torch
-
-pipeline_text2img = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True
-)
-print(type(pipeline_text2img))
-"<class 'diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline'>"
-```
-
-Then [`~AutoPipelineForImage2Image.from_pipe`] maps the original `"stable-diffusion"` pipeline class to [`StableDiffusionImg2ImgPipeline`]:
-
-```py
-pipeline_img2img = AutoPipelineForImage2Image.from_pipe(pipeline_text2img)
-print(type(pipeline_img2img))
-"<class 'diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline'>"
-```
-
-If you passed an optional argument - like disabling the safety checker - to the original pipeline, this argument is also passed on to the new pipeline:
-
-```py
-from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
-import torch
-
-pipeline_text2img = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-    requires_safety_checker=False,
-).to("cuda")
-
-pipeline_img2img = AutoPipelineForImage2Image.from_pipe(pipeline_text2img)
-print(pipeline_img2img.config.requires_safety_checker)
-"False"
-```
-
-You can overwrite any of the arguments and even configuration from the original pipeline if you want to change the behavior of the new pipeline. For example, to turn the safety checker back on and add the `strength` argument:
-
-```py
-pipeline_img2img = AutoPipelineForImage2Image.from_pipe(pipeline_text2img, requires_safety_checker=True, strength=0.3)
-print(pipeline_img2img.config.requires_safety_checker)
-"True"
-```
--- a/docs/source/en/using-diffusers/ip_adapter.md
+++ b/docs/source/en/using-diffusers/ip_adapter.md
@@ -640,3 +640,87 @@ image
 <div class="flex justify-center">
     <img src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/ipa-controlnet-out.png" />
 </div>
+
+### Style & layout control
+
+[InstantStyle](https://arxiv.org/abs/2404.02733) is a plug-and-play method on top of IP-Adapter, which disentangles style and layout from image prompt to control image generation. This is achieved by only inserting IP-Adapters to some specific part of the model. 
+
+By default IP-Adapters are inserted to all layers of the model. Use the [`~loaders.IPAdapterMixin.set_ip_adapter_scale`] method with a dictionary to assign scales to IP-Adapter at different layers.
+
+```py
+from diffusers import AutoPipelineForImage2Image
+from diffusers.utils import load_image
+import torch
+
+pipeline = AutoPipelineForImage2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
+pipeline.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter_sdxl.bin")
+
+scale = {
+    "down": {"block_2": [0.0, 1.0]},
+    "up": {"block_0": [0.0, 1.0, 0.0]},
+}
+pipeline.set_ip_adapter_scale(scale)
+```
+
+This will activate IP-Adapter at the second layer in the model's down-part block 2 and up-part block 0. The former is the layer where IP-Adapter injects layout information and the latter injects style. Inserting IP-Adapter to these two layers you can generate images following the style and layout of image prompt, but with contents more aligned to text prompt.
+
+```py
+style_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg")
+
+generator = torch.Generator(device="cpu").manual_seed(42)
+image = pipeline(
+    prompt="a cat, masterpiece, best quality, high quality",
+    image=style_image,
+    negative_prompt="text, watermark, lowres, low quality, worst quality, deformed, glitch, low contrast, noisy, saturation, blurry",
+    guidance_scale=5,
+    num_inference_steps=30,
+    generator=generator,
+).images[0]
+image
+```
+
+<div class="flex flex-row gap-4">
+  <div class="flex-1">
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">IP-Adapter image</figcaption>
+  </div>
+  <div class="flex-1">
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit_style_layout_cat.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
+  </div>
+</div>
+
+In contrast, inserting IP-Adapter to all layers will often generate images that overly focus on image prompt and diminish diversity.
+
+Activate IP-Adapter only in the style layer and then call the pipeline again.
+
+```py
+scale = {
+    "up": {"block_0": [0.0, 1.0, 0.0]},
+}
+pipeline.set_ip_adapter_scale(scale)
+
+generator = torch.Generator(device="cpu").manual_seed(42)
+image = pipeline(
+    prompt="a cat, masterpiece, best quality, high quality",
+    image=style_image,
+    negative_prompt="text, watermark, lowres, low quality, worst quality, deformed, glitch, low contrast, noisy, saturation, blurry",
+    guidance_scale=5,
+    num_inference_steps=30,
+    generator=generator,
+).images[0]
+image
+```
+
+<div class="flex flex-row gap-4">
+  <div class="flex-1">
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit_style_cat.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">IP-Adapter only in style layer</figcaption>
+  </div>
+  <div class="flex-1">
+    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/30518dfe089e6bf50008875077b44cb98fb2065c/diffusers/default_out.png"/>
+    <figcaption class="mt-2 text-center text-sm text-gray-500">IP-Adapter in all layers</figcaption>
+  </div>
+</div>
+
+Note that you don't have to specify all layers in the dictionary. Those not included in the dictionary will be set to scale 0 which means disable IP-Adapter by default.
--- a/docs/source/en/using-diffusers/loading.md
+++ b/docs/source/en/using-diffusers/loading.md
@@ -154,7 +154,10 @@ When you load multiple pipelines that share the same model components, it makes
 1. You generated an image with the [`StableDiffusionPipeline`] but you want to improve its quality with the [`StableDiffusionSAGPipeline`]. Both of these pipelines share the same pretrained model, so it'd be a waste of memory to load the same model twice.
 2. You want to add a model component, like a [`MotionAdapter`](../api/pipelines/animatediff#animatediffpipeline), to [`AnimateDiffPipeline`] which was instantiated from an existing [`StableDiffusionPipeline`]. Again, both pipelines share the same pretrained model, so it'd be a waste of memory to load an entirely new pipeline again.

-With the [`DiffusionPipeline.from_pipe`] API, you can switch between multiple pipelines to take advantage of their different features without increasing memory-usage. It is similar to turning on and off a feature in your pipeline. To switch between tasks, use the [`~DiffusionPipeline.from_pipe`] method with the [`AutoPipeline`](../api/pipelines/auto_pipeline) class, which automatically identifies the pipeline class based on the task (learn more in the [AutoPipeline](../tutorials/autopipeline) tutorial).
+With the [`DiffusionPipeline.from_pipe`] API, you can switch between multiple pipelines to take advantage of their different features without increasing memory-usage. It is similar to turning on and off a feature in your pipeline.
+
+> [!TIP]
+> To switch between tasks (rather than features), use the [`~DiffusionPipeline.from_pipe`] method with the [AutoPipeline](../api/pipelines/auto_pipeline) class, which automatically identifies the pipeline class based on the task (learn more in the [AutoPipeline](../tutorials/autopipeline) tutorial).

 Let's start with a [`StableDiffusionPipeline`] and then reuse the loaded model components to create a [`StableDiffusionSAGPipeline`] to increase generation quality. You'll use the [`StableDiffusionPipeline`] with an [IP-Adapter](./ip_adapter) to generate a bear eating pizza.

--- a/docs/source/en/using-diffusers/other-modalities.md
+++ b/docs/source/en/using-diffusers/other-modalities.md
@@ -1,21 +0,0 @@
-<!--Copyright 2024 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. 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.
-->
-
-# Using Diffusers with other modalities
-
-Diffusers is in the process of expanding to modalities other than images.
-
-Example type        | Colab | Pipeline |
-:-------------------------:|:-------------------------:|:-------------------------:|
-[Molecule conformation](https://www.nature.com/subjects/molecular-conformation#:~:text=Definition,to%20changes%20in%20their%20environment.) generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/geodiff_molecule_conformation.ipynb) | ❌
-
-More coming soon!
--- a/docs/source/en/using-diffusers/overview_techniques.md
+++ b/docs/source/en/using-diffusers/overview_techniques.md
@@ -0,0 +1,18 @@
+<!--Copyright 2024 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. 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.
+-->
+
+# Overview
+
+The inference pipeline supports and enables a wide range of techniques that are divided into two categories:
+
+* Pipeline functionality: these techniques modify the pipeline or extend it for other applications. For example, pipeline callbacks add new features to a pipeline and a pipeline can also be extended for distributed inference.
+* Improve inference quality: these techniques increase the visual quality of the generated images. For example, you can enhance your prompts with GPT2 to create better images with lower effort.
--- a/docs/source/en/using-diffusers/pipeline_overview.md
+++ b/docs/source/en/using-diffusers/pipeline_overview.md
@@ -1,17 +0,0 @@
-<!--Copyright 2024 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. 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.
-->
-
-# Overview
-
-A pipeline is an end-to-end class that provides a quick and easy way to use a diffusion system for inference by bundling independently trained models and schedulers together. Certain combinations of models and schedulers define specific pipeline types, like [`StableDiffusionXLPipeline`] or [`StableDiffusionControlNetPipeline`], with specific capabilities. All pipeline types inherit from the base [`DiffusionPipeline`] class; pass it any checkpoint, and it'll automatically detect the pipeline type and load the necessary components.
-
-This section demonstrates how to use specific pipelines such as Stable Diffusion XL, ControlNet, and DiffEdit. You'll also learn how to use a distilled version of the Stable Diffusion model to speed up inference, how to create reproducible pipelines, and how to use and contribute community pipelines.
--- a/examples/research_projects/geodiff/README.md
+++ b/examples/research_projects/geodiff/README.md
@@ -0,0 +1,6 @@
+# GeoDiff
+
+> [!TIP]
+> This notebook is not actively maintained by the Diffusers team. For any questions or comments, please contact [natolambert](https://twitter.com/natolambert).
+
+This is an experimental research notebook demonstrating how to generate stable 3D structures of molecules with [GeoDiff](https://github.com/MinkaiXu/GeoDiff) and Diffusers.
--- a/examples/research_projects/geodiff/geodiff_molecule_conformation.ipynb
+++ b/examples/research_projects/geodiff/geodiff_molecule_conformation.ipynb
--- a/scripts/convert_pixart_sigma_to_diffusers.py
+++ b/scripts/convert_pixart_sigma_to_diffusers.py
@@ -0,0 +1,223 @@
+import argparse
+import os
+
+import torch
+from transformers import T5EncoderModel, T5Tokenizer
+
+from diffusers import AutoencoderKL, DPMSolverMultistepScheduler, PixArtSigmaPipeline, Transformer2DModel
+
+
+ckpt_id = "PixArt-alpha"
+# https://github.com/PixArt-alpha/PixArt-sigma/blob/dd087141864e30ec44f12cb7448dd654be065e88/scripts/inference.py#L158
+interpolation_scale = {256: 0.5, 512: 1, 1024: 2, 2048: 4}
+
+
+def main(args):
+    all_state_dict = torch.load(args.orig_ckpt_path)
+    state_dict = all_state_dict.pop("state_dict")
+    converted_state_dict = {}
+
+    # Patch embeddings.
+    converted_state_dict["pos_embed.proj.weight"] = state_dict.pop("x_embedder.proj.weight")
+    converted_state_dict["pos_embed.proj.bias"] = state_dict.pop("x_embedder.proj.bias")
+
+    # Caption projection.
+    converted_state_dict["caption_projection.linear_1.weight"] = state_dict.pop("y_embedder.y_proj.fc1.weight")
+    converted_state_dict["caption_projection.linear_1.bias"] = state_dict.pop("y_embedder.y_proj.fc1.bias")
+    converted_state_dict["caption_projection.linear_2.weight"] = state_dict.pop("y_embedder.y_proj.fc2.weight")
+    converted_state_dict["caption_projection.linear_2.bias"] = state_dict.pop("y_embedder.y_proj.fc2.bias")
+
+    # AdaLN-single LN
+    converted_state_dict["adaln_single.emb.timestep_embedder.linear_1.weight"] = state_dict.pop(
+        "t_embedder.mlp.0.weight"
+    )
+    converted_state_dict["adaln_single.emb.timestep_embedder.linear_1.bias"] = state_dict.pop("t_embedder.mlp.0.bias")
+    converted_state_dict["adaln_single.emb.timestep_embedder.linear_2.weight"] = state_dict.pop(
+        "t_embedder.mlp.2.weight"
+    )
+    converted_state_dict["adaln_single.emb.timestep_embedder.linear_2.bias"] = state_dict.pop("t_embedder.mlp.2.bias")
+
+    if args.micro_condition:
+        # Resolution.
+        converted_state_dict["adaln_single.emb.resolution_embedder.linear_1.weight"] = state_dict.pop(
+            "csize_embedder.mlp.0.weight"
+        )
+        converted_state_dict["adaln_single.emb.resolution_embedder.linear_1.bias"] = state_dict.pop(
+            "csize_embedder.mlp.0.bias"
+        )
+        converted_state_dict["adaln_single.emb.resolution_embedder.linear_2.weight"] = state_dict.pop(
+            "csize_embedder.mlp.2.weight"
+        )
+        converted_state_dict["adaln_single.emb.resolution_embedder.linear_2.bias"] = state_dict.pop(
+            "csize_embedder.mlp.2.bias"
+        )
+        # Aspect ratio.
+        converted_state_dict["adaln_single.emb.aspect_ratio_embedder.linear_1.weight"] = state_dict.pop(
+            "ar_embedder.mlp.0.weight"
+        )
+        converted_state_dict["adaln_single.emb.aspect_ratio_embedder.linear_1.bias"] = state_dict.pop(
+            "ar_embedder.mlp.0.bias"
+        )
+        converted_state_dict["adaln_single.emb.aspect_ratio_embedder.linear_2.weight"] = state_dict.pop(
+            "ar_embedder.mlp.2.weight"
+        )
+        converted_state_dict["adaln_single.emb.aspect_ratio_embedder.linear_2.bias"] = state_dict.pop(
+            "ar_embedder.mlp.2.bias"
+        )
+    # Shared norm.
+    converted_state_dict["adaln_single.linear.weight"] = state_dict.pop("t_block.1.weight")
+    converted_state_dict["adaln_single.linear.bias"] = state_dict.pop("t_block.1.bias")
+
+    for depth in range(28):
+        # Transformer blocks.
+        converted_state_dict[f"transformer_blocks.{depth}.scale_shift_table"] = state_dict.pop(
+            f"blocks.{depth}.scale_shift_table"
+        )
+        # Attention is all you need 🤘
+
+        # Self attention.
+        q, k, v = torch.chunk(state_dict.pop(f"blocks.{depth}.attn.qkv.weight"), 3, dim=0)
+        q_bias, k_bias, v_bias = torch.chunk(state_dict.pop(f"blocks.{depth}.attn.qkv.bias"), 3, dim=0)
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_q.weight"] = q
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_q.bias"] = q_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_k.weight"] = k
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_k.bias"] = k_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_v.weight"] = v
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_v.bias"] = v_bias
+        # Projection.
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_out.0.weight"] = state_dict.pop(
+            f"blocks.{depth}.attn.proj.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_out.0.bias"] = state_dict.pop(
+            f"blocks.{depth}.attn.proj.bias"
+        )
+        if args.qk_norm:
+            converted_state_dict[f"transformer_blocks.{depth}.attn1.q_norm.weight"] = state_dict.pop(
+                f"blocks.{depth}.attn.q_norm.weight"
+            )
+            converted_state_dict[f"transformer_blocks.{depth}.attn1.q_norm.bias"] = state_dict.pop(
+                f"blocks.{depth}.attn.q_norm.bias"
+            )
+            converted_state_dict[f"transformer_blocks.{depth}.attn1.k_norm.weight"] = state_dict.pop(
+                f"blocks.{depth}.attn.k_norm.weight"
+            )
+            converted_state_dict[f"transformer_blocks.{depth}.attn1.k_norm.bias"] = state_dict.pop(
+                f"blocks.{depth}.attn.k_norm.bias"
+            )
+
+        # Feed-forward.
+        converted_state_dict[f"transformer_blocks.{depth}.ff.net.0.proj.weight"] = state_dict.pop(
+            f"blocks.{depth}.mlp.fc1.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.net.0.proj.bias"] = state_dict.pop(
+            f"blocks.{depth}.mlp.fc1.bias"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.net.2.weight"] = state_dict.pop(
+            f"blocks.{depth}.mlp.fc2.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.net.2.bias"] = state_dict.pop(
+            f"blocks.{depth}.mlp.fc2.bias"
+        )
+
+        # Cross-attention.
+        q = state_dict.pop(f"blocks.{depth}.cross_attn.q_linear.weight")
+        q_bias = state_dict.pop(f"blocks.{depth}.cross_attn.q_linear.bias")
+        k, v = torch.chunk(state_dict.pop(f"blocks.{depth}.cross_attn.kv_linear.weight"), 2, dim=0)
+        k_bias, v_bias = torch.chunk(state_dict.pop(f"blocks.{depth}.cross_attn.kv_linear.bias"), 2, dim=0)
+
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_q.weight"] = q
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_q.bias"] = q_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_k.weight"] = k
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_k.bias"] = k_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_v.weight"] = v
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_v.bias"] = v_bias
+
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_out.0.weight"] = state_dict.pop(
+            f"blocks.{depth}.cross_attn.proj.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_out.0.bias"] = state_dict.pop(
+            f"blocks.{depth}.cross_attn.proj.bias"
+        )
+
+    # Final block.
+    converted_state_dict["proj_out.weight"] = state_dict.pop("final_layer.linear.weight")
+    converted_state_dict["proj_out.bias"] = state_dict.pop("final_layer.linear.bias")
+    converted_state_dict["scale_shift_table"] = state_dict.pop("final_layer.scale_shift_table")
+
+    # PixArt XL/2
+    transformer = Transformer2DModel(
+        sample_size=args.image_size // 8,
+        num_layers=28,
+        attention_head_dim=72,
+        in_channels=4,
+        out_channels=8,
+        patch_size=2,
+        attention_bias=True,
+        num_attention_heads=16,
+        cross_attention_dim=1152,
+        activation_fn="gelu-approximate",
+        num_embeds_ada_norm=1000,
+        norm_type="ada_norm_single",
+        norm_elementwise_affine=False,
+        norm_eps=1e-6,
+        caption_channels=4096,
+        interpolation_scale=interpolation_scale[args.image_size],
+        use_additional_conditions=args.micro_condition,
+    )
+    transformer.load_state_dict(converted_state_dict, strict=True)
+
+    assert transformer.pos_embed.pos_embed is not None
+    try:
+        state_dict.pop("y_embedder.y_embedding")
+        state_dict.pop("pos_embed")
+    except Exception as e:
+        print(f"Skipping {str(e)}")
+        pass
+    assert len(state_dict) == 0, f"State dict is not empty, {state_dict.keys()}"
+
+    num_model_params = sum(p.numel() for p in transformer.parameters())
+    print(f"Total number of transformer parameters: {num_model_params}")
+
+    if args.only_transformer:
+        transformer.save_pretrained(os.path.join(args.dump_path, "transformer"))
+    else:
+        # pixart-Sigma vae link: https://huggingface.co/PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers/tree/main/vae
+        vae = AutoencoderKL.from_pretrained(f"{ckpt_id}/pixart_sigma_sdxlvae_T5_diffusers", subfolder="vae")
+
+        scheduler = DPMSolverMultistepScheduler()
+
+        tokenizer = T5Tokenizer.from_pretrained(f"{ckpt_id}/pixart_sigma_sdxlvae_T5_diffusers", subfolder="tokenizer")
+        text_encoder = T5EncoderModel.from_pretrained(
+            f"{ckpt_id}/pixart_sigma_sdxlvae_T5_diffusers", subfolder="text_encoder"
+        )
+
+        pipeline = PixArtSigmaPipeline(
+            tokenizer=tokenizer, text_encoder=text_encoder, transformer=transformer, vae=vae, scheduler=scheduler
+        )
+
+        pipeline.save_pretrained(args.dump_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--micro_condition", action="store_true", help="If use Micro-condition in PixArtMS structure during training."
+    )
+    parser.add_argument("--qk_norm", action="store_true", help="If use qk norm during training.")
+    parser.add_argument(
+        "--orig_ckpt_path", default=None, type=str, required=False, help="Path to the checkpoint to convert."
+    )
+    parser.add_argument(
+        "--image_size",
+        default=1024,
+        type=int,
+        choices=[256, 512, 1024, 2048],
+        required=False,
+        help="Image size of pretrained model, 256, 512, 1024, or 2048.",
+    )
+    parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output pipeline.")
+    parser.add_argument("--only_transformer", default=True, type=bool, required=True)
+
+    args = parser.parse_args()
+    main(args)
--- a/src/diffusers/init.py
+++ b/src/diffusers/init.py
@@ -261,6 +261,7 @@ else:
            "PaintByExamplePipeline",
            "PIAPipeline",
            "PixArtAlphaPipeline",
+            "PixArtSigmaPipeline",
            "SemanticStableDiffusionPipeline",
            "ShapEImg2ImgPipeline",
            "ShapEPipeline",
@@ -637,6 +638,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
            PaintByExamplePipeline,
            PIAPipeline,
            PixArtAlphaPipeline,
+            PixArtSigmaPipeline,
            SemanticStableDiffusionPipeline,
            ShapEImg2ImgPipeline,
            ShapEPipeline,
--- a/src/diffusers/configuration_utils.py
+++ b/src/diffusers/configuration_utils.py
@@ -450,8 +450,8 @@ class ConfigMixin:
        return outputs

    @staticmethod
-    def _get_init_keys(cls):
-        return set(dict(inspect.signature(cls.__init__).parameters).keys())
+    def _get_init_keys(input_class):
+        return set(dict(inspect.signature(input_class.__init__).parameters).keys())

    @classmethod
    def extract_init_dict(cls, config_dict, **kwargs):
--- a/src/diffusers/image_processor.py
+++ b/src/diffusers/image_processor.py
@@ -994,3 +994,77 @@ class IPAdapterMaskProcessor(VaeImageProcessor):
        )

        return mask_downsample
+
+
+class PixArtImageProcessor(VaeImageProcessor):
+    """
+    Image processor for PixArt image resize and crop.
+
+    Args:
+        do_resize (`bool`, *optional*, defaults to `True`):
+            Whether to downscale the image's (height, width) dimensions to multiples of `vae_scale_factor`. Can accept
+            `height` and `width` arguments from [`image_processor.VaeImageProcessor.preprocess`] method.
+        vae_scale_factor (`int`, *optional*, defaults to `8`):
+            VAE scale factor. If `do_resize` is `True`, the image is automatically resized to multiples of this factor.
+        resample (`str`, *optional*, defaults to `lanczos`):
+            Resampling filter to use when resizing the image.
+        do_normalize (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the image to [-1,1].
+        do_binarize (`bool`, *optional*, defaults to `False`):
+            Whether to binarize the image to 0/1.
+        do_convert_rgb (`bool`, *optional*, defaults to be `False`):
+            Whether to convert the images to RGB format.
+        do_convert_grayscale (`bool`, *optional*, defaults to be `False`):
+            Whether to convert the images to grayscale format.
+    """
+
+    @register_to_config
+    def __init__(
+        self,
+        do_resize: bool = True,
+        vae_scale_factor: int = 8,
+        resample: str = "lanczos",
+        do_normalize: bool = True,
+        do_binarize: bool = False,
+        do_convert_grayscale: bool = False,
+    ):
+        super().__init__(
+            do_resize=do_resize,
+            vae_scale_factor=vae_scale_factor,
+            resample=resample,
+            do_normalize=do_normalize,
+            do_binarize=do_binarize,
+            do_convert_grayscale=do_convert_grayscale,
+        )
+
+    @staticmethod
+    def classify_height_width_bin(height: int, width: int, ratios: dict) -> Tuple[int, int]:
+        """Returns binned height and width."""
+        ar = float(height / width)
+        closest_ratio = min(ratios.keys(), key=lambda ratio: abs(float(ratio) - ar))
+        default_hw = ratios[closest_ratio]
+        return int(default_hw[0]), int(default_hw[1])
+
+    @staticmethod
+    def resize_and_crop_tensor(samples: torch.Tensor, new_width: int, new_height: int) -> torch.Tensor:
+        orig_height, orig_width = samples.shape[2], samples.shape[3]
+
+        # Check if resizing is needed
+        if orig_height != new_height or orig_width != new_width:
+            ratio = max(new_height / orig_height, new_width / orig_width)
+            resized_width = int(orig_width * ratio)
+            resized_height = int(orig_height * ratio)
+
+            # Resize
+            samples = F.interpolate(
+                samples, size=(resized_height, resized_width), mode="bilinear", align_corners=False
+            )
+
+            # Center Crop
+            start_x = (resized_width - new_width) // 2
+            end_x = start_x + new_width
+            start_y = (resized_height - new_height) // 2
+            end_y = start_y + new_height
+            samples = samples[:, :, start_y:end_y, start_x:end_x]
+
+        return samples
--- a/src/diffusers/loaders/ip_adapter.py
+++ b/src/diffusers/loaders/ip_adapter.py
@@ -16,6 +16,7 @@ from pathlib import Path
 from typing import Dict, List, Optional, Union

 import torch
+import torch.nn.functional as F
 from huggingface_hub.utils import validate_hf_hub_args
 from safetensors import safe_open

@@ -28,6 +29,7 @@ from ..utils import (
    is_transformers_available,
    logging,
 )
+from .unet_loader_utils import _maybe_expand_lora_scales


 if is_transformers_available():
@@ -37,6 +39,8 @@ if is_transformers_available():
    )

    from ..models.attention_processor import (
+        AttnProcessor,
+        AttnProcessor2_0,
        IPAdapterAttnProcessor,
        IPAdapterAttnProcessor2_0,
    )
@@ -243,25 +247,55 @@ class IPAdapterMixin:

    def set_ip_adapter_scale(self, scale):
        """
-        Sets the conditioning scale between text and image.
+        Set IP-Adapter scales per-transformer block. Input `scale` could be a single config or a list of configs for
+        granular control over each IP-Adapter behavior. A config can be a float or a dictionary.

        Example:

        ```py
-        pipeline.set_ip_adapter_scale(0.5)
+        # To use original IP-Adapter
+        scale = 1.0
+        pipeline.set_ip_adapter_scale(scale)
+
+        # To use style block only
+        scale = {
+            "up": {"block_0": [0.0, 1.0, 0.0]},
+        }
+        pipeline.set_ip_adapter_scale(scale)
+
+        # To use style+layout blocks
+        scale = {
+            "down": {"block_2": [0.0, 1.0]},
+            "up": {"block_0": [0.0, 1.0, 0.0]},
+        }
+        pipeline.set_ip_adapter_scale(scale)
+
+        # To use style and layout from 2 reference images
+        scales = [{"down": {"block_2": [0.0, 1.0]}}, {"up": {"block_0": [0.0, 1.0, 0.0]}}]
+        pipeline.set_ip_adapter_scale(scales)
        ```
        """
        unet = getattr(self, self.unet_name) if not hasattr(self, "unet") else self.unet
-        for attn_processor in unet.attn_processors.values():
+        if not isinstance(scale, list):
+            scale = [scale]
+        scale_configs = _maybe_expand_lora_scales(unet, scale, default_scale=0.0)
+
+        for attn_name, attn_processor in unet.attn_processors.items():
            if isinstance(attn_processor, (IPAdapterAttnProcessor, IPAdapterAttnProcessor2_0)):
-                if not isinstance(scale, list):
-                    scale = [scale] * len(attn_processor.scale)
-                if len(attn_processor.scale) != len(scale):
+                if len(scale_configs) != len(attn_processor.scale):
                    raise ValueError(
-                        f"`scale` should be a list of same length as the number if ip-adapters "
-                        f"Expected {len(attn_processor.scale)} but got {len(scale)}."
+                        f"Cannot assign {len(scale_configs)} scale_configs to "
+                        f"{len(attn_processor.scale)} IP-Adapter."
                    )
-                attn_processor.scale = scale
+                elif len(scale_configs) == 1:
+                    scale_configs = scale_configs * len(attn_processor.scale)
+                for i, scale_config in enumerate(scale_configs):
+                    if isinstance(scale_config, dict):
+                        for k, s in scale_config.items():
+                            if attn_name.startswith(k):
+                                attn_processor.scale[i] = s
+                    else:
+                        attn_processor.scale[i] = scale_config

    def unload_ip_adapter(self):
        """
@@ -292,4 +326,14 @@ class IPAdapterMixin:
        self.config.encoder_hid_dim_type = None

        # restore original Unet attention processors layers
-        self.unet.set_default_attn_processor()
+        attn_procs = {}
+        for name, value in self.unet.attn_processors.items():
+            attn_processor_class = (
+                AttnProcessor2_0() if hasattr(F, "scaled_dot_product_attention") else AttnProcessor()
+            )
+            attn_procs[name] = (
+                attn_processor_class
+                if isinstance(value, (IPAdapterAttnProcessor, IPAdapterAttnProcessor2_0))
+                else value.__class__()
+            )
+        self.unet.set_attn_processor(attn_procs)
--- a/src/diffusers/loaders/unet_loader_utils.py
+++ b/src/diffusers/loaders/unet_loader_utils.py
@@ -38,7 +38,9 @@ def _translate_into_actual_layer_name(name):
    return ".".join((updown, block, attn))


-def _maybe_expand_lora_scales(unet: "UNet2DConditionModel", weight_scales: List[Union[float, Dict]]):
+def _maybe_expand_lora_scales(
+    unet: "UNet2DConditionModel", weight_scales: List[Union[float, Dict]], default_scale=1.0
+):
    blocks_with_transformer = {
        "down": [i for i, block in enumerate(unet.down_blocks) if hasattr(block, "attentions")],
        "up": [i for i, block in enumerate(unet.up_blocks) if hasattr(block, "attentions")],
@@ -47,7 +49,11 @@ def _maybe_expand_lora_scales(unet: "UNet2DConditionModel", weight_scales: List[

    expanded_weight_scales = [
        _maybe_expand_lora_scales_for_one_adapter(
-            weight_for_adapter, blocks_with_transformer, transformer_per_block, unet.state_dict()
+            weight_for_adapter,
+            blocks_with_transformer,
+            transformer_per_block,
+            unet.state_dict(),
+            default_scale=default_scale,
        )
        for weight_for_adapter in weight_scales
    ]
@@ -60,6 +66,7 @@ def _maybe_expand_lora_scales_for_one_adapter(
    blocks_with_transformer: Dict[str, int],
    transformer_per_block: Dict[str, int],
    state_dict: None,
+    default_scale: float = 1.0,
 ):
    """
    Expands the inputs into a more granular dictionary. See the example below for more details.
@@ -108,21 +115,36 @@ def _maybe_expand_lora_scales_for_one_adapter(
    scales = copy.deepcopy(scales)

    if "mid" not in scales:
-        scales["mid"] = 1
+        scales["mid"] = default_scale
+    elif isinstance(scales["mid"], list):
+        if len(scales["mid"]) == 1:
+            scales["mid"] = scales["mid"][0]
+        else:
+            raise ValueError(f"Expected 1 scales for mid, got {len(scales['mid'])}.")

    for updown in ["up", "down"]:
        if updown not in scales:
-            scales[updown] = 1
+            scales[updown] = default_scale

        # eg {"down": 1} to {"down": {"block_1": 1, "block_2": 1}}}
        if not isinstance(scales[updown], dict):
-            scales[updown] = {f"block_{i}": scales[updown] for i in blocks_with_transformer[updown]}
+            scales[updown] = {f"block_{i}": copy.deepcopy(scales[updown]) for i in blocks_with_transformer[updown]}

-        # eg {"down": "block_1": 1}} to {"down": "block_1": [1, 1]}}
+        # eg {"down": {"block_1": 1}} to {"down": {"block_1": [1, 1]}}
        for i in blocks_with_transformer[updown]:
            block = f"block_{i}"
+            # set not assigned blocks to default scale
+            if block not in scales[updown]:
+                scales[updown][block] = default_scale
            if not isinstance(scales[updown][block], list):
                scales[updown][block] = [scales[updown][block] for _ in range(transformer_per_block[updown])]
+            elif len(scales[updown][block]) == 1:
+                # a list specifying scale to each masked IP input
+                scales[updown][block] = scales[updown][block] * transformer_per_block[updown]
+            elif len(scales[updown][block]) != transformer_per_block[updown]:
+                raise ValueError(
+                    f"Expected {transformer_per_block[updown]} scales for {updown}.{block}, got {len(scales[updown][block])}."
+                )

        # eg {"down": "block_1": [1, 1]}}  to {"down.block_1.0": 1, "down.block_1.1": 1}
        for i in blocks_with_transformer[updown]:
--- a/src/diffusers/models/attention_processor.py
+++ b/src/diffusers/models/attention_processor.py
@@ -2229,44 +2229,51 @@ class IPAdapterAttnProcessor(nn.Module):
        for current_ip_hidden_states, scale, to_k_ip, to_v_ip, mask in zip(
            ip_hidden_states, self.scale, self.to_k_ip, self.to_v_ip, ip_adapter_masks
        ):
-            if mask is not None:
-                if not isinstance(scale, list):
-                    scale = [scale]
+            skip = False
+            if isinstance(scale, list):
+                if all(s == 0 for s in scale):
+                    skip = True
+            elif scale == 0:
+                skip = True
+            if not skip:
+                if mask is not None:
+                    if not isinstance(scale, list):
+                        scale = [scale] * mask.shape[1]

-                current_num_images = mask.shape[1]
-                for i in range(current_num_images):
-                    ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
-                    ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])
+                    current_num_images = mask.shape[1]
+                    for i in range(current_num_images):
+                        ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
+                        ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])
+
+                        ip_key = attn.head_to_batch_dim(ip_key)
+                        ip_value = attn.head_to_batch_dim(ip_value)
+
+                        ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
+                        _current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
+                        _current_ip_hidden_states = attn.batch_to_head_dim(_current_ip_hidden_states)
+
+                        mask_downsample = IPAdapterMaskProcessor.downsample(
+                            mask[:, i, :, :],
+                            batch_size,
+                            _current_ip_hidden_states.shape[1],
+                            _current_ip_hidden_states.shape[2],
+                        )
+
+                        mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)
+
+                        hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
+                else:
+                    ip_key = to_k_ip(current_ip_hidden_states)
+                    ip_value = to_v_ip(current_ip_hidden_states)

                    ip_key = attn.head_to_batch_dim(ip_key)
                    ip_value = attn.head_to_batch_dim(ip_value)

                    ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
-                    _current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
-                    _current_ip_hidden_states = attn.batch_to_head_dim(_current_ip_hidden_states)
+                    current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
+                    current_ip_hidden_states = attn.batch_to_head_dim(current_ip_hidden_states)

-                    mask_downsample = IPAdapterMaskProcessor.downsample(
-                        mask[:, i, :, :],
-                        batch_size,
-                        _current_ip_hidden_states.shape[1],
-                        _current_ip_hidden_states.shape[2],
-                    )
-
-                    mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)
-
-                    hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
-            else:
-                ip_key = to_k_ip(current_ip_hidden_states)
-                ip_value = to_v_ip(current_ip_hidden_states)
-
-                ip_key = attn.head_to_batch_dim(ip_key)
-                ip_value = attn.head_to_batch_dim(ip_value)
-
-                ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
-                current_ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
-                current_ip_hidden_states = attn.batch_to_head_dim(current_ip_hidden_states)
-
-                hidden_states = hidden_states + scale * current_ip_hidden_states
+                    hidden_states = hidden_states + scale * current_ip_hidden_states

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
@@ -2439,57 +2446,64 @@ class IPAdapterAttnProcessor2_0(torch.nn.Module):
        for current_ip_hidden_states, scale, to_k_ip, to_v_ip, mask in zip(
            ip_hidden_states, self.scale, self.to_k_ip, self.to_v_ip, ip_adapter_masks
        ):
-            if mask is not None:
-                if not isinstance(scale, list):
-                    scale = [scale]
+            skip = False
+            if isinstance(scale, list):
+                if all(s == 0 for s in scale):
+                    skip = True
+            elif scale == 0:
+                skip = True
+            if not skip:
+                if mask is not None:
+                    if not isinstance(scale, list):
+                        scale = [scale] * mask.shape[1]

-                current_num_images = mask.shape[1]
-                for i in range(current_num_images):
-                    ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
-                    ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])
+                    current_num_images = mask.shape[1]
+                    for i in range(current_num_images):
+                        ip_key = to_k_ip(current_ip_hidden_states[:, i, :, :])
+                        ip_value = to_v_ip(current_ip_hidden_states[:, i, :, :])
+
+                        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+                        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+                        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+                        # TODO: add support for attn.scale when we move to Torch 2.1
+                        _current_ip_hidden_states = F.scaled_dot_product_attention(
+                            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
+                        )
+
+                        _current_ip_hidden_states = _current_ip_hidden_states.transpose(1, 2).reshape(
+                            batch_size, -1, attn.heads * head_dim
+                        )
+                        _current_ip_hidden_states = _current_ip_hidden_states.to(query.dtype)
+
+                        mask_downsample = IPAdapterMaskProcessor.downsample(
+                            mask[:, i, :, :],
+                            batch_size,
+                            _current_ip_hidden_states.shape[1],
+                            _current_ip_hidden_states.shape[2],
+                        )
+
+                        mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)
+                        hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
+                else:
+                    ip_key = to_k_ip(current_ip_hidden_states)
+                    ip_value = to_v_ip(current_ip_hidden_states)

                    ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
                    ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

                    # the output of sdp = (batch, num_heads, seq_len, head_dim)
                    # TODO: add support for attn.scale when we move to Torch 2.1
-                    _current_ip_hidden_states = F.scaled_dot_product_attention(
+                    current_ip_hidden_states = F.scaled_dot_product_attention(
                        query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
                    )

-                    _current_ip_hidden_states = _current_ip_hidden_states.transpose(1, 2).reshape(
+                    current_ip_hidden_states = current_ip_hidden_states.transpose(1, 2).reshape(
                        batch_size, -1, attn.heads * head_dim
                    )
-                    _current_ip_hidden_states = _current_ip_hidden_states.to(query.dtype)
+                    current_ip_hidden_states = current_ip_hidden_states.to(query.dtype)

-                    mask_downsample = IPAdapterMaskProcessor.downsample(
-                        mask[:, i, :, :],
-                        batch_size,
-                        _current_ip_hidden_states.shape[1],
-                        _current_ip_hidden_states.shape[2],
-                    )
-
-                    mask_downsample = mask_downsample.to(dtype=query.dtype, device=query.device)
-                    hidden_states = hidden_states + scale[i] * (_current_ip_hidden_states * mask_downsample)
-            else:
-                ip_key = to_k_ip(current_ip_hidden_states)
-                ip_value = to_v_ip(current_ip_hidden_states)
-
-                ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-                ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
-                # the output of sdp = (batch, num_heads, seq_len, head_dim)
-                # TODO: add support for attn.scale when we move to Torch 2.1
-                current_ip_hidden_states = F.scaled_dot_product_attention(
-                    query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
-                )
-
-                current_ip_hidden_states = current_ip_hidden_states.transpose(1, 2).reshape(
-                    batch_size, -1, attn.heads * head_dim
-                )
-                current_ip_hidden_states = current_ip_hidden_states.to(query.dtype)
-
-                hidden_states = hidden_states + scale * current_ip_hidden_states
+                    hidden_states = hidden_states + scale * current_ip_hidden_states

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
--- a/src/diffusers/models/autoencoders/vae.py
+++ b/src/diffusers/models/autoencoders/vae.py
@@ -90,7 +90,6 @@ class Encoder(nn.Module):
            padding=1,
        )

-        self.mid_block = None
        self.down_blocks = nn.ModuleList([])

        # down
@@ -228,7 +227,6 @@ class Decoder(nn.Module):
            padding=1,
        )

-        self.mid_block = None
        self.up_blocks = nn.ModuleList([])

        temb_channels = in_channels if norm_type == "spatial" else None
@@ -474,7 +472,6 @@ class MaskConditionDecoder(nn.Module):
            padding=1,
        )

-        self.mid_block = None
        self.up_blocks = nn.ModuleList([])

        temb_channels = in_channels if norm_type == "spatial" else None
--- a/src/diffusers/models/transformers/transformer_2d.py
+++ b/src/diffusers/models/transformers/transformer_2d.py
@@ -100,6 +100,7 @@ class Transformer2DModel(ModelMixin, ConfigMixin):
        attention_type: str = "default",
        caption_channels: int = None,
        interpolation_scale: float = None,
+        use_additional_conditions: Optional[bool] = None,
    ):
        super().__init__()

@@ -124,6 +125,12 @@ class Transformer2DModel(ModelMixin, ConfigMixin):
        self.in_channels = in_channels
        self.out_channels = in_channels if out_channels is None else out_channels
        self.gradient_checkpointing = False
+        if use_additional_conditions is None:
+            if norm_type == "ada_norm_single" and sample_size == 128:
+                use_additional_conditions = True
+            else:
+                use_additional_conditions = False
+        self.use_additional_conditions = use_additional_conditions

        # 1. Transformer2DModel can process both standard continuous images of shape `(batch_size, num_channels, width, height)` as well as quantized image embeddings of shape `(batch_size, num_image_vectors)`
        # Define whether input is continuous or discrete depending on configuration
@@ -305,9 +312,7 @@ class Transformer2DModel(ModelMixin, ConfigMixin):

        # PixArt-Alpha blocks.
        self.adaln_single = None
-        self.use_additional_conditions = False
        if self.config.norm_type == "ada_norm_single":
-            self.use_additional_conditions = self.config.sample_size == 128
            # TODO(Sayak, PVP) clean this, for now we use sample size to determine whether to use
            # additional conditions until we find better name
            self.adaln_single = AdaLayerNormSingle(
--- a/src/diffusers/pipelines/init.py
+++ b/src/diffusers/pipelines/init.py
@@ -187,7 +187,7 @@ else:
    _import_structure["musicldm"] = ["MusicLDMPipeline"]
    _import_structure["paint_by_example"] = ["PaintByExamplePipeline"]
    _import_structure["pia"] = ["PIAPipeline"]
-    _import_structure["pixart_alpha"] = ["PixArtAlphaPipeline"]
+    _import_structure["pixart_alpha"] = ["PixArtAlphaPipeline", "PixArtSigmaPipeline"]
    _import_structure["semantic_stable_diffusion"] = ["SemanticStableDiffusionPipeline"]
    _import_structure["shap_e"] = ["ShapEImg2ImgPipeline", "ShapEPipeline"]
    _import_structure["stable_cascade"] = [
@@ -450,7 +450,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
        from .musicldm import MusicLDMPipeline
        from .paint_by_example import PaintByExamplePipeline
        from .pia import PIAPipeline
-        from .pixart_alpha import PixArtAlphaPipeline
+        from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
        from .semantic_stable_diffusion import SemanticStableDiffusionPipeline
        from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline
        from .stable_cascade import (
--- a/src/diffusers/pipelines/pixart_alpha/init.py
+++ b/src/diffusers/pipelines/pixart_alpha/init.py
@@ -23,6 +23,7 @@ except OptionalDependencyNotAvailable:
    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
    _import_structure["pipeline_pixart_alpha"] = ["PixArtAlphaPipeline"]
+    _import_structure["pipeline_pixart_sigma"] = ["PixArtSigmaPipeline"]

 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
    try:
@@ -32,7 +33,13 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
    except OptionalDependencyNotAvailable:
        from ...utils.dummy_torch_and_transformers_objects import *
    else:
-        from .pipeline_pixart_alpha import PixArtAlphaPipeline
+        from .pipeline_pixart_alpha import (
+            ASPECT_RATIO_256_BIN,
+            ASPECT_RATIO_512_BIN,
+            ASPECT_RATIO_1024_BIN,
+            PixArtAlphaPipeline,
+        )
+        from .pipeline_pixart_sigma import ASPECT_RATIO_2048_BIN, PixArtSigmaPipeline

 else:
    import sys
--- a/src/diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py
+++ b/src/diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py
@@ -19,10 +19,9 @@ import urllib.parse as ul
 from typing import Callable, List, Optional, Tuple, Union

 import torch
-import torch.nn.functional as F
 from transformers import T5EncoderModel, T5Tokenizer

-from ...image_processor import VaeImageProcessor
+from ...image_processor import PixArtImageProcessor
 from ...models import AutoencoderKL, Transformer2DModel
 from ...schedulers import DPMSolverMultistepScheduler
 from ...utils import (
@@ -272,16 +271,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
        )

        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
-        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
-
-    # Adapted from https://github.com/PixArt-alpha/PixArt-alpha/blob/master/diffusion/model/utils.py
-    def mask_text_embeddings(self, emb, mask):
-        if emb.shape[0] == 1:
-            keep_index = mask.sum().item()
-            return emb[:, :, :keep_index, :], keep_index
-        else:
-            masked_feature = emb * mask[:, None, :, None]
-            return masked_feature, emb.shape[2]
+        self.image_processor = PixArtImageProcessor(vae_scale_factor=self.vae_scale_factor)

    # Adapted from diffusers.pipelines.deepfloyd_if.pipeline_if.encode_prompt
    def encode_prompt(
@@ -674,38 +664,6 @@ class PixArtAlphaPipeline(DiffusionPipeline):
        latents = latents * self.scheduler.init_noise_sigma
        return latents

-    @staticmethod
-    def classify_height_width_bin(height: int, width: int, ratios: dict) -> Tuple[int, int]:
-        """Returns binned height and width."""
-        ar = float(height / width)
-        closest_ratio = min(ratios.keys(), key=lambda ratio: abs(float(ratio) - ar))
-        default_hw = ratios[closest_ratio]
-        return int(default_hw[0]), int(default_hw[1])
-
-    @staticmethod
-    def resize_and_crop_tensor(samples: torch.Tensor, new_width: int, new_height: int) -> torch.Tensor:
-        orig_height, orig_width = samples.shape[2], samples.shape[3]
-
-        # Check if resizing is needed
-        if orig_height != new_height or orig_width != new_width:
-            ratio = max(new_height / orig_height, new_width / orig_width)
-            resized_width = int(orig_width * ratio)
-            resized_height = int(orig_height * ratio)
-
-            # Resize
-            samples = F.interpolate(
-                samples, size=(resized_height, resized_width), mode="bilinear", align_corners=False
-            )
-
-            # Center Crop
-            start_x = (resized_width - new_width) // 2
-            end_x = start_x + new_width
-            start_y = (resized_height - new_height) // 2
-            end_y = start_y + new_height
-            samples = samples[:, :, start_y:end_y, start_x:end_x]
-
-        return samples
-
    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def __call__(
@@ -826,7 +784,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
            else:
                raise ValueError("Invalid sample size")
            orig_height, orig_width = height, width
-            height, width = self.classify_height_width_bin(height, width, ratios=aspect_ratio_bin)
+            height, width = self.image_processor.classify_height_width_bin(height, width, ratios=aspect_ratio_bin)

        self.check_inputs(
            prompt,
@@ -956,7 +914,11 @@ class PixArtAlphaPipeline(DiffusionPipeline):
                    noise_pred = noise_pred

                # compute previous image: x_t -> x_t-1
-                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                if num_inference_steps == 1:
+                    # For DMD one step sampling: https://arxiv.org/abs/2311.18828
+                    latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).pred_original_sample
+                else:
+                    latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -968,7 +930,7 @@ class PixArtAlphaPipeline(DiffusionPipeline):
        if not output_type == "latent":
            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
            if use_resolution_binning:
-                image = self.resize_and_crop_tensor(image, orig_width, orig_height)
+                image = self.image_processor.resize_and_crop_tensor(image, orig_width, orig_height)
        else:
            image = latents

--- a/src/diffusers/pipelines/pixart_alpha/pipeline_pixart_sigma.py
+++ b/src/diffusers/pipelines/pixart_alpha/pipeline_pixart_sigma.py
@@ -0,0 +1,857 @@
+# Copyright 2024 PixArt-Sigma Authors and 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.
+# 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 html
+import inspect
+import re
+import urllib.parse as ul
+from typing import Callable, List, Optional, Tuple, Union
+
+import torch
+from transformers import T5EncoderModel, T5Tokenizer
+
+from ...image_processor import PixArtImageProcessor
+from ...models import AutoencoderKL, Transformer2DModel
+from ...schedulers import DPMSolverMultistepScheduler
+from ...utils import (
+    BACKENDS_MAPPING,
+    deprecate,
+    is_bs4_available,
+    is_ftfy_available,
+    logging,
+    replace_example_docstring,
+)
+from ...utils.torch_utils import randn_tensor
+from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from .pipeline_pixart_alpha import (
+    ASPECT_RATIO_256_BIN,
+    ASPECT_RATIO_512_BIN,
+    ASPECT_RATIO_1024_BIN,
+)
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+if is_bs4_available():
+    from bs4 import BeautifulSoup
+
+if is_ftfy_available():
+    import ftfy
+
+
+ASPECT_RATIO_2048_BIN = {
+    "0.25": [1024.0, 4096.0],
+    "0.26": [1024.0, 3968.0],
+    "0.27": [1024.0, 3840.0],
+    "0.28": [1024.0, 3712.0],
+    "0.32": [1152.0, 3584.0],
+    "0.33": [1152.0, 3456.0],
+    "0.35": [1152.0, 3328.0],
+    "0.4": [1280.0, 3200.0],
+    "0.42": [1280.0, 3072.0],
+    "0.48": [1408.0, 2944.0],
+    "0.5": [1408.0, 2816.0],
+    "0.52": [1408.0, 2688.0],
+    "0.57": [1536.0, 2688.0],
+    "0.6": [1536.0, 2560.0],
+    "0.68": [1664.0, 2432.0],
+    "0.72": [1664.0, 2304.0],
+    "0.78": [1792.0, 2304.0],
+    "0.82": [1792.0, 2176.0],
+    "0.88": [1920.0, 2176.0],
+    "0.94": [1920.0, 2048.0],
+    "1.0": [2048.0, 2048.0],
+    "1.07": [2048.0, 1920.0],
+    "1.13": [2176.0, 1920.0],
+    "1.21": [2176.0, 1792.0],
+    "1.29": [2304.0, 1792.0],
+    "1.38": [2304.0, 1664.0],
+    "1.46": [2432.0, 1664.0],
+    "1.67": [2560.0, 1536.0],
+    "1.75": [2688.0, 1536.0],
+    "2.0": [2816.0, 1408.0],
+    "2.09": [2944.0, 1408.0],
+    "2.4": [3072.0, 1280.0],
+    "2.5": [3200.0, 1280.0],
+    "2.89": [3328.0, 1152.0],
+    "3.0": [3456.0, 1152.0],
+    "3.11": [3584.0, 1152.0],
+    "3.62": [3712.0, 1024.0],
+    "3.75": [3840.0, 1024.0],
+    "3.88": [3968.0, 1024.0],
+    "4.0": [4096.0, 1024.0],
+}
+
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import PixArtSigmaPipeline
+
+        >>> # You can replace the checkpoint id with "PixArt-alpha/PixArt-Sigma-XL-2-512-MS" too.
+        >>> pipe = PixArtSigmaPipeline.from_pretrained(
+        ...     "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS", torch_dtype=torch.float16
+        ... )
+        >>> # Enable memory optimizations.
+        >>> # pipe.enable_model_cpu_offload()
+
+        >>> prompt = "A small cactus with a happy face in the Sahara desert."
+        >>> image = pipe(prompt).images[0]
+        ```
+"""
+
+
+# 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,
+    **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 support arbitrary spacing between timesteps. If `None`, then the default
+                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
+                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:
+        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)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class PixArtSigmaPipeline(DiffusionPipeline):
+    r"""
+    Pipeline for text-to-image generation using PixArt-Sigma.
+    """
+
+    bad_punct_regex = re.compile(
+        r"["
+        + "#®•©™&@·º½¾¿¡§~"
+        + r"\)"
+        + r"\("
+        + r"\]"
+        + r"\["
+        + r"\}"
+        + r"\{"
+        + r"\|"
+        + "\\"
+        + r"\/"
+        + r"\*"
+        + r"]{1,}"
+    )  # noqa
+
+    _optional_components = ["tokenizer", "text_encoder"]
+    model_cpu_offload_seq = "text_encoder->transformer->vae"
+
+    def __init__(
+        self,
+        tokenizer: T5Tokenizer,
+        text_encoder: T5EncoderModel,
+        vae: AutoencoderKL,
+        transformer: Transformer2DModel,
+        scheduler: DPMSolverMultistepScheduler,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
+        )
+
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.image_processor = PixArtImageProcessor(vae_scale_factor=self.vae_scale_factor)
+
+    # Copied from diffusers.pipelines.pixart_alpha.pipeline_pixart_alpha.PixArtAlphaPipeline.encode_prompt
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        do_classifier_free_guidance: bool = True,
+        negative_prompt: str = "",
+        num_images_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_attention_mask: Optional[torch.FloatTensor] = None,
+        negative_prompt_attention_mask: Optional[torch.FloatTensor] = None,
+        clean_caption: bool = False,
+        max_sequence_length: int = 120,
+        **kwargs,
+    ):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
+                instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). For
+                PixArt-Alpha, this should be "".
+            do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+                whether to use classifier free guidance or not
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                number of images that should be generated per prompt
+            device: (`torch.device`, *optional*):
+                torch device to place the resulting embeddings on
+            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.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. For PixArt-Alpha, it's should be the embeddings of the ""
+                string.
+            clean_caption (`bool`, defaults to `False`):
+                If `True`, the function will preprocess and clean the provided caption before encoding.
+            max_sequence_length (`int`, defaults to 120): Maximum sequence length to use for the prompt.
+        """
+
+        if "mask_feature" in kwargs:
+            deprecation_message = "The use of `mask_feature` is deprecated. It is no longer used in any computation and that doesn't affect the end results. It will be removed in a future version."
+            deprecate("mask_feature", "1.0.0", deprecation_message, standard_warn=False)
+
+        if device is None:
+            device = self._execution_device
+
+        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]
+
+        # See Section 3.1. of the paper.
+        max_length = max_sequence_length
+
+        if prompt_embeds is None:
+            prompt = self._text_preprocessing(prompt, clean_caption=clean_caption)
+            text_inputs = self.tokenizer(
+                prompt,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                add_special_tokens=True,
+                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[:, max_length - 1 : -1])
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {max_length} tokens: {removed_text}"
+                )
+
+            prompt_attention_mask = text_inputs.attention_mask
+            prompt_attention_mask = prompt_attention_mask.to(device)
+
+            prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=prompt_attention_mask)
+            prompt_embeds = prompt_embeds[0]
+
+        if self.text_encoder is not None:
+            dtype = self.text_encoder.dtype
+        elif self.transformer is not None:
+            dtype = self.transformer.dtype
+        else:
+            dtype = None
+
+        prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+        bs_embed, 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(bs_embed * num_images_per_prompt, seq_len, -1)
+        prompt_attention_mask = prompt_attention_mask.view(bs_embed, -1)
+        prompt_attention_mask = prompt_attention_mask.repeat(num_images_per_prompt, 1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance and negative_prompt_embeds is None:
+            uncond_tokens = [negative_prompt] * batch_size
+            uncond_tokens = self._text_preprocessing(uncond_tokens, clean_caption=clean_caption)
+            max_length = prompt_embeds.shape[1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_attention_mask=True,
+                add_special_tokens=True,
+                return_tensors="pt",
+            )
+            negative_prompt_attention_mask = uncond_input.attention_mask
+            negative_prompt_attention_mask = negative_prompt_attention_mask.to(device)
+
+            negative_prompt_embeds = self.text_encoder(
+                uncond_input.input_ids.to(device), attention_mask=negative_prompt_attention_mask
+            )
+            negative_prompt_embeds = negative_prompt_embeds[0]
+
+        if do_classifier_free_guidance:
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = negative_prompt_embeds.shape[1]
+
+            negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device)
+
+            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+            negative_prompt_attention_mask = negative_prompt_attention_mask.view(bs_embed, -1)
+            negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(num_images_per_prompt, 1)
+        else:
+            negative_prompt_embeds = None
+            negative_prompt_attention_mask = None
+
+        return prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.pixart_alpha.pipeline_pixart_alpha.PixArtAlphaPipeline.check_inputs
+    def check_inputs(
+        self,
+        prompt,
+        height,
+        width,
+        negative_prompt,
+        callback_steps,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        prompt_attention_mask=None,
+        negative_prompt_attention_mask=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_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+        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 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)}")
+
+        if prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and prompt_attention_mask is None:
+            raise ValueError("Must provide `prompt_attention_mask` when specifying `prompt_embeds`.")
+
+        if negative_prompt_embeds is not None and negative_prompt_attention_mask is None:
+            raise ValueError("Must provide `negative_prompt_attention_mask` when specifying `negative_prompt_embeds`.")
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+            if prompt_attention_mask.shape != negative_prompt_attention_mask.shape:
+                raise ValueError(
+                    "`prompt_attention_mask` and `negative_prompt_attention_mask` must have the same shape when passed directly, but"
+                    f" got: `prompt_attention_mask` {prompt_attention_mask.shape} != `negative_prompt_attention_mask`"
+                    f" {negative_prompt_attention_mask.shape}."
+                )
+
+    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline._text_preprocessing
+    def _text_preprocessing(self, text, clean_caption=False):
+        if clean_caption and not is_bs4_available():
+            logger.warning(BACKENDS_MAPPING["bs4"][-1].format("Setting `clean_caption=True`"))
+            logger.warning("Setting `clean_caption` to False...")
+            clean_caption = False
+
+        if clean_caption and not is_ftfy_available():
+            logger.warning(BACKENDS_MAPPING["ftfy"][-1].format("Setting `clean_caption=True`"))
+            logger.warning("Setting `clean_caption` to False...")
+            clean_caption = False
+
+        if not isinstance(text, (tuple, list)):
+            text = [text]
+
+        def process(text: str):
+            if clean_caption:
+                text = self._clean_caption(text)
+                text = self._clean_caption(text)
+            else:
+                text = text.lower().strip()
+            return text
+
+        return [process(t) for t in text]
+
+    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline._clean_caption
+    def _clean_caption(self, caption):
+        caption = str(caption)
+        caption = ul.unquote_plus(caption)
+        caption = caption.strip().lower()
+        caption = re.sub("<person>", "person", caption)
+        # urls:
+        caption = re.sub(
+            r"\b((?:https?:(?:\/{1,3}|[a-zA-Z0-9%])|[a-zA-Z0-9.\-]+[.](?:com|co|ru|net|org|edu|gov|it)[\w/-]*\b\/?(?!@)))",  # noqa
+            "",
+            caption,
+        )  # regex for urls
+        caption = re.sub(
+            r"\b((?:www:(?:\/{1,3}|[a-zA-Z0-9%])|[a-zA-Z0-9.\-]+[.](?:com|co|ru|net|org|edu|gov|it)[\w/-]*\b\/?(?!@)))",  # noqa
+            "",
+            caption,
+        )  # regex for urls
+        # html:
+        caption = BeautifulSoup(caption, features="html.parser").text
+
+        # @<nickname>
+        caption = re.sub(r"@[\w\d]+\b", "", caption)
+
+        # 31C0—31EF CJK Strokes
+        # 31F0—31FF Katakana Phonetic Extensions
+        # 3200—32FF Enclosed CJK Letters and Months
+        # 3300—33FF CJK Compatibility
+        # 3400—4DBF CJK Unified Ideographs Extension A
+        # 4DC0—4DFF Yijing Hexagram Symbols
+        # 4E00—9FFF CJK Unified Ideographs
+        caption = re.sub(r"[\u31c0-\u31ef]+", "", caption)
+        caption = re.sub(r"[\u31f0-\u31ff]+", "", caption)
+        caption = re.sub(r"[\u3200-\u32ff]+", "", caption)
+        caption = re.sub(r"[\u3300-\u33ff]+", "", caption)
+        caption = re.sub(r"[\u3400-\u4dbf]+", "", caption)
+        caption = re.sub(r"[\u4dc0-\u4dff]+", "", caption)
+        caption = re.sub(r"[\u4e00-\u9fff]+", "", caption)
+        #######################################################
+
+        # все виды тире / all types of dash --> "-"
+        caption = re.sub(
+            r"[\u002D\u058A\u05BE\u1400\u1806\u2010-\u2015\u2E17\u2E1A\u2E3A\u2E3B\u2E40\u301C\u3030\u30A0\uFE31\uFE32\uFE58\uFE63\uFF0D]+",  # noqa
+            "-",
+            caption,
+        )
+
+        # кавычки к одному стандарту
+        caption = re.sub(r"[`´«»“”¨]", '"', caption)
+        caption = re.sub(r"[‘’]", "'", caption)
+
+        # &quot;
+        caption = re.sub(r"&quot;?", "", caption)
+        # &amp
+        caption = re.sub(r"&amp", "", caption)
+
+        # ip adresses:
+        caption = re.sub(r"\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}", " ", caption)
+
+        # article ids:
+        caption = re.sub(r"\d:\d\d\s+$", "", caption)
+
+        # \n
+        caption = re.sub(r"\\n", " ", caption)
+
+        # "#123"
+        caption = re.sub(r"#\d{1,3}\b", "", caption)
+        # "#12345.."
+        caption = re.sub(r"#\d{5,}\b", "", caption)
+        # "123456.."
+        caption = re.sub(r"\b\d{6,}\b", "", caption)
+        # filenames:
+        caption = re.sub(r"[\S]+\.(?:png|jpg|jpeg|bmp|webp|eps|pdf|apk|mp4)", "", caption)
+
+        #
+        caption = re.sub(r"[\"\']{2,}", r'"', caption)  # """AUSVERKAUFT"""
+        caption = re.sub(r"[\.]{2,}", r" ", caption)  # """AUSVERKAUFT"""
+
+        caption = re.sub(self.bad_punct_regex, r" ", caption)  # ***AUSVERKAUFT***, #AUSVERKAUFT
+        caption = re.sub(r"\s+\.\s+", r" ", caption)  # " . "
+
+        # this-is-my-cute-cat / this_is_my_cute_cat
+        regex2 = re.compile(r"(?:\-|\_)")
+        if len(re.findall(regex2, caption)) > 3:
+            caption = re.sub(regex2, " ", caption)
+
+        caption = ftfy.fix_text(caption)
+        caption = html.unescape(html.unescape(caption))
+
+        caption = re.sub(r"\b[a-zA-Z]{1,3}\d{3,15}\b", "", caption)  # jc6640
+        caption = re.sub(r"\b[a-zA-Z]+\d+[a-zA-Z]+\b", "", caption)  # jc6640vc
+        caption = re.sub(r"\b\d+[a-zA-Z]+\d+\b", "", caption)  # 6640vc231
+
+        caption = re.sub(r"(worldwide\s+)?(free\s+)?shipping", "", caption)
+        caption = re.sub(r"(free\s)?download(\sfree)?", "", caption)
+        caption = re.sub(r"\bclick\b\s(?:for|on)\s\w+", "", caption)
+        caption = re.sub(r"\b(?:png|jpg|jpeg|bmp|webp|eps|pdf|apk|mp4)(\simage[s]?)?", "", caption)
+        caption = re.sub(r"\bpage\s+\d+\b", "", caption)
+
+        caption = re.sub(r"\b\d*[a-zA-Z]+\d+[a-zA-Z]+\d+[a-zA-Z\d]*\b", r" ", caption)  # j2d1a2a...
+
+        caption = re.sub(r"\b\d+\.?\d*[xх×]\d+\.?\d*\b", "", caption)
+
+        caption = re.sub(r"\b\s+\:\s+", r": ", caption)
+        caption = re.sub(r"(\D[,\./])\b", r"\1 ", caption)
+        caption = re.sub(r"\s+", " ", caption)
+
+        caption.strip()
+
+        caption = re.sub(r"^[\"\']([\w\W]+)[\"\']$", r"\1", caption)
+        caption = re.sub(r"^[\'\_,\-\:;]", r"", caption)
+        caption = re.sub(r"[\'\_,\-\:\-\+]$", r"", caption)
+        caption = re.sub(r"^\.\S+$", "", caption)
+
+        return caption.strip()
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (
+            batch_size,
+            num_channels_latents,
+            int(height) // self.vae_scale_factor,
+            int(width) // self.vae_scale_factor,
+        )
+        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."
+            )
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        negative_prompt: str = "",
+        num_inference_steps: int = 20,
+        timesteps: List[int] = None,
+        guidance_scale: float = 4.5,
+        num_images_per_prompt: Optional[int] = 1,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        prompt_attention_mask: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_attention_mask: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        clean_caption: bool = True,
+        use_resolution_binning: bool = True,
+        max_sequence_length: int = 300,
+        **kwargs,
+    ) -> Union[ImagePipelineOutput, Tuple]:
+        """
+        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.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+                less than `1`).
+            num_inference_steps (`int`, *optional*, defaults to 100):
+                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. If not defined, equal spaced `num_inference_steps`
+                timesteps are used. Must be in descending order.
+            guidance_scale (`float`, *optional*, defaults to 4.5):
+                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.
+            height (`int`, *optional*, defaults to self.unet.config.sample_size):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size):
+                The width in pixels of the generated image.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            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.
+            prompt_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask for text embeddings.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. For PixArt-Sigma this negative prompt should be "". If not
+                provided, negative_prompt_embeds will be generated from `negative_prompt` input argument.
+            negative_prompt_attention_mask (`torch.FloatTensor`, *optional*):
+                Pre-generated attention mask for negative text embeddings.
+            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.stable_diffusion.IFPipelineOutput`] instead of a plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+            clean_caption (`bool`, *optional*, defaults to `True`):
+                Whether or not to clean the caption before creating embeddings. Requires `beautifulsoup4` and `ftfy` to
+                be installed. If the dependencies are not installed, the embeddings will be created from the raw
+                prompt.
+            use_resolution_binning (`bool` defaults to `True`):
+                If set to `True`, the requested height and width are first mapped to the closest resolutions using
+                `ASPECT_RATIO_1024_BIN`. After the produced latents are decoded into images, they are resized back to
+                the requested resolution. Useful for generating non-square images.
+            max_sequence_length (`int` defaults to 120): Maximum sequence length to use with the `prompt`.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.ImagePipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is
+                returned where the first element is a list with the generated images
+        """
+        # 1. Check inputs. Raise error if not correct
+        height = height or self.transformer.config.sample_size * self.vae_scale_factor
+        width = width or self.transformer.config.sample_size * self.vae_scale_factor
+        if use_resolution_binning:
+            if self.transformer.config.sample_size == 256:
+                aspect_ratio_bin = ASPECT_RATIO_2048_BIN
+            elif self.transformer.config.sample_size == 128:
+                aspect_ratio_bin = ASPECT_RATIO_1024_BIN
+            elif self.transformer.config.sample_size == 64:
+                aspect_ratio_bin = ASPECT_RATIO_512_BIN
+            elif self.transformer.config.sample_size == 32:
+                aspect_ratio_bin = ASPECT_RATIO_256_BIN
+            else:
+                raise ValueError("Invalid sample size")
+            orig_height, orig_width = height, width
+            height, width = self.image_processor.classify_height_width_bin(height, width, ratios=aspect_ratio_bin)
+
+        self.check_inputs(
+            prompt,
+            height,
+            width,
+            negative_prompt,
+            callback_steps,
+            prompt_embeds,
+            negative_prompt_embeds,
+            prompt_attention_mask,
+            negative_prompt_attention_mask,
+        )
+
+        # 2. Default height and width to transformer
+        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
+
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        (
+            prompt_embeds,
+            prompt_attention_mask,
+            negative_prompt_embeds,
+            negative_prompt_attention_mask,
+        ) = self.encode_prompt(
+            prompt,
+            do_classifier_free_guidance,
+            negative_prompt=negative_prompt,
+            num_images_per_prompt=num_images_per_prompt,
+            device=device,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            negative_prompt_attention_mask=negative_prompt_attention_mask,
+            clean_caption=clean_caption,
+            max_sequence_length=max_sequence_length,
+        )
+        if do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+            prompt_attention_mask = torch.cat([negative_prompt_attention_mask, prompt_attention_mask], dim=0)
+
+        # 4. Prepare timesteps
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+
+        # 5. Prepare latents.
+        latent_channels = self.transformer.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            latent_channels,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 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)
+
+        # 6.1 Prepare micro-conditions.
+        added_cond_kwargs = {"resolution": None, "aspect_ratio": None}
+
+        # 7. 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:
+            for i, t in enumerate(timesteps):
+                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+                current_timestep = t
+                if not torch.is_tensor(current_timestep):
+                    # TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
+                    # This would be a good case for the `match` statement (Python 3.10+)
+                    is_mps = latent_model_input.device.type == "mps"
+                    if isinstance(current_timestep, float):
+                        dtype = torch.float32 if is_mps else torch.float64
+                    else:
+                        dtype = torch.int32 if is_mps else torch.int64
+                    current_timestep = torch.tensor([current_timestep], dtype=dtype, device=latent_model_input.device)
+                elif len(current_timestep.shape) == 0:
+                    current_timestep = current_timestep[None].to(latent_model_input.device)
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                current_timestep = current_timestep.expand(latent_model_input.shape[0])
+
+                # predict noise model_output
+                noise_pred = self.transformer(
+                    latent_model_input,
+                    encoder_hidden_states=prompt_embeds,
+                    encoder_attention_mask=prompt_attention_mask,
+                    timestep=current_timestep,
+                    added_cond_kwargs=added_cond_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # perform guidance
+                if do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                # learned sigma
+                if self.transformer.config.out_channels // 2 == latent_channels:
+                    noise_pred = noise_pred.chunk(2, dim=1)[0]
+                else:
+                    noise_pred = noise_pred
+
+                # compute previous image: x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                # 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 callback is not None and i % callback_steps == 0:
+                        step_idx = i // getattr(self.scheduler, "order", 1)
+                        callback(step_idx, t, latents)
+
+        if not output_type == "latent":
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+            if use_resolution_binning:
+                image = self.image_processor.resize_and_crop_tensor(image, orig_width, orig_height)
+        else:
+            image = latents
+
+        if not output_type == "latent":
+            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 ImagePipelineOutput(images=image)
--- a/src/diffusers/utils/dummy_torch_and_transformers_objects.py
+++ b/src/diffusers/utils/dummy_torch_and_transformers_objects.py
@@ -737,6 +737,21 @@ class PixArtAlphaPipeline(metaclass=DummyObject):
        requires_backends(cls, ["torch", "transformers"])


+class PixArtSigmaPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class SemanticStableDiffusionPipeline(metaclass=DummyObject):
    _backends = ["torch", "transformers"]

--- a/tests/models/autoencoders/test_models_vae.py
+++ b/tests/models/autoencoders/test_models_vae.py
@@ -1153,5 +1153,5 @@ class ConsistencyDecoderVAEIntegrationTests(unittest.TestCase):
        shapes = [(1, 4, 73, 97), (1, 4, 97, 73), (1, 4, 49, 65), (1, 4, 65, 49)]
        with torch.no_grad():
            for shape in shapes:
-                image = torch.zeros(shape, device=torch_device)
+                image = torch.zeros(shape, device=torch_device, dtype=pipe.vae.dtype)
                pipe.vae.decode(image)
--- a/tests/models/unets/test_models_unet_controlnetxs.py
+++ b/tests/models/unets/test_models_unet_controlnetxs.py
@@ -22,11 +22,7 @@ from torch import nn

 from diffusers import ControlNetXSAdapter, UNet2DConditionModel, UNetControlNetXSModel
 from diffusers.utils import logging
-from diffusers.utils.testing_utils import (
-    enable_full_determinism,
-    floats_tensor,
-    torch_device,
-)
+from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, is_flaky, torch_device

 from ..test_modeling_common import ModelTesterMixin, UNetTesterMixin

@@ -305,6 +301,7 @@ class UNetControlNetXSModelTests(ModelTesterMixin, UNetTesterMixin, unittest.Tes
        assert set(modules_with_gc_enabled.keys()) == EXPECTED_SET
        assert all(modules_with_gc_enabled.values()), "All modules should be enabled"

+    @is_flaky
    def test_forward_no_control(self):
        unet = self.get_dummy_unet()
        controlnet = self.get_dummy_controlnet_from_unet(unet)
--- a/tests/pipelines/ip_adapters/test_ip_adapter_stable_diffusion.py
+++ b/tests/pipelines/ip_adapters/test_ip_adapter_stable_diffusion.py
@@ -32,7 +32,6 @@ from diffusers import (
    StableDiffusionXLPipeline,
 )
 from diffusers.image_processor import IPAdapterMaskProcessor
-from diffusers.models.attention_processor import AttnProcessor, AttnProcessor2_0
 from diffusers.utils import load_image
 from diffusers.utils.testing_utils import (
    enable_full_determinism,
@@ -73,7 +72,9 @@ class IPAdapterNightlyTestsMixin(unittest.TestCase):
        image_processor = CLIPImageProcessor.from_pretrained(repo_id)
        return image_processor

-    def get_dummy_inputs(self, for_image_to_image=False, for_inpainting=False, for_sdxl=False, for_masks=False):
+    def get_dummy_inputs(
+        self, for_image_to_image=False, for_inpainting=False, for_sdxl=False, for_masks=False, for_instant_style=False
+    ):
        image = load_image(
            "https://user-images.githubusercontent.com/24734142/266492875-2d50d223-8475-44f0-a7c6-08b51cb53572.png"
        )
@@ -126,6 +127,40 @@ class IPAdapterNightlyTestsMixin(unittest.TestCase):
                }
            )

+        elif for_instant_style:
+            composition_mask = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/1024_whole_mask.png"
+            )
+            female_mask = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter_None_20240321125641_mask.png"
+            )
+            male_mask = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter_None_20240321125344_mask.png"
+            )
+            background_mask = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter_6_20240321130722_mask.png"
+            )
+            ip_composition_image = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter__20240321125152.png"
+            )
+            ip_female_style = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter__20240321125625.png"
+            )
+            ip_male_style = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter__20240321125329.png"
+            )
+            ip_background = load_image(
+                "https://huggingface.co/datasets/OzzyGT/testing-resources/resolve/main/ip_adapter__20240321130643.png"
+            )
+            input_kwargs.update(
+                {
+                    "ip_adapter_image": [ip_composition_image, [ip_female_style, ip_male_style, ip_background]],
+                    "cross_attention_kwargs": {
+                        "ip_adapter_masks": [[composition_mask], [female_mask, male_mask, background_mask]]
+                    },
+                }
+            )
+
        return input_kwargs


@@ -271,6 +306,7 @@ class IPAdapterSDIntegrationTests(IPAdapterNightlyTestsMixin):
        pipeline = StableDiffusionPipeline.from_pretrained(
            "runwayml/stable-diffusion-v1-5", image_encoder=image_encoder, safety_checker=None, torch_dtype=self.dtype
        )
+        before_processors = [attn_proc.__class__ for attn_proc in pipeline.unet.attn_processors.values()]
        pipeline.to(torch_device)
        pipeline.load_ip_adapter("h94/IP-Adapter", subfolder="models", weight_name="ip-adapter_sd15.bin")
        pipeline.set_ip_adapter_scale(0.7)
@@ -279,11 +315,9 @@ class IPAdapterSDIntegrationTests(IPAdapterNightlyTestsMixin):

        assert getattr(pipeline, "image_encoder") is None
        assert getattr(pipeline, "feature_extractor") is not None
-        processors = [
-            isinstance(attn_proc, (AttnProcessor, AttnProcessor2_0))
-            for name, attn_proc in pipeline.unet.attn_processors.items()
-        ]
-        assert processors == [True] * len(processors)
+        after_processors = [attn_proc.__class__ for attn_proc in pipeline.unet.attn_processors.values()]
+
+        assert before_processors == after_processors

    @is_flaky
    def test_multi(self):
@@ -575,6 +609,48 @@ class IPAdapterSDXLIntegrationTests(IPAdapterNightlyTestsMixin):
        max_diff = numpy_cosine_similarity_distance(image_slice, expected_slice)
        assert max_diff < 5e-4

+    def test_instant_style_multiple_masks(self):
+        image_encoder = CLIPVisionModelWithProjection.from_pretrained(
+            "h94/IP-Adapter", subfolder="models/image_encoder", torch_dtype=torch.float16
+        ).to("cuda")
+        pipeline = StableDiffusionXLPipeline.from_pretrained(
+            "RunDiffusion/Juggernaut-XL-v9", torch_dtype=torch.float16, image_encoder=image_encoder, variant="fp16"
+        ).to("cuda")
+        pipeline.enable_model_cpu_offload()
+
+        pipeline.load_ip_adapter(
+            ["ostris/ip-composition-adapter", "h94/IP-Adapter"],
+            subfolder=["", "sdxl_models"],
+            weight_name=[
+                "ip_plus_composition_sdxl.safetensors",
+                "ip-adapter_sdxl_vit-h.safetensors",
+            ],
+            image_encoder_folder=None,
+        )
+        scale_1 = {
+            "down": [[0.0, 0.0, 1.0]],
+            "mid": [[0.0, 0.0, 1.0]],
+            "up": {"block_0": [[0.0, 0.0, 1.0], [1.0, 1.0, 1.0], [0.0, 0.0, 1.0]], "block_1": [[0.0, 0.0, 1.0]]},
+        }
+        pipeline.set_ip_adapter_scale([1.0, scale_1])
+
+        inputs = self.get_dummy_inputs(for_instant_style=True)
+        processor = IPAdapterMaskProcessor()
+        masks1 = inputs["cross_attention_kwargs"]["ip_adapter_masks"][0]
+        masks2 = inputs["cross_attention_kwargs"]["ip_adapter_masks"][1]
+        masks1 = processor.preprocess(masks1, height=1024, width=1024)
+        masks2 = processor.preprocess(masks2, height=1024, width=1024)
+        masks2 = masks2.reshape(1, masks2.shape[0], masks2.shape[2], masks2.shape[3])
+        inputs["cross_attention_kwargs"]["ip_adapter_masks"] = [masks1, masks2]
+        images = pipeline(**inputs).images
+        image_slice = images[0, :3, :3, -1].flatten()
+        expected_slice = np.array(
+            [0.23551631, 0.20476806, 0.14099443, 0.0, 0.07675594, 0.05672678, 0.0, 0.0, 0.02099729]
+        )
+
+        max_diff = numpy_cosine_similarity_distance(image_slice, expected_slice)
+        assert max_diff < 5e-4
+
    def test_ip_adapter_multiple_masks_one_adapter(self):
        image_encoder = self.get_image_encoder(repo_id="h94/IP-Adapter", subfolder="models/image_encoder")
        pipeline = StableDiffusionXLPipeline.from_pretrained(
--- a/tests/pipelines/kandinsky/test_kandinsky.py
+++ b/tests/pipelines/kandinsky/test_kandinsky.py
@@ -299,7 +299,7 @@ class KandinskyPipelineIntegrationTests(unittest.TestCase):
        pipe_prior.to(torch_device)

        pipeline = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
-        pipeline = pipeline.to(torch_device)
+        pipeline.to(torch_device)
        pipeline.set_progress_bar_config(disable=None)

        prompt = "red cat, 4k photo"
--- a/tests/pipelines/kandinsky2_2/test_kandinsky.py
+++ b/tests/pipelines/kandinsky2_2/test_kandinsky.py
@@ -25,11 +25,12 @@ from diffusers.utils.testing_utils import (
    enable_full_determinism,
    floats_tensor,
    load_numpy,
+    numpy_cosine_similarity_distance,
    require_torch_gpu,
    slow,
 )

-from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
+from ..test_pipelines_common import PipelineTesterMixin


 enable_full_determinism()
@@ -248,12 +249,12 @@ class KandinskyV22PipelineIntegrationTests(unittest.TestCase):
        pipeline = KandinskyV22Pipeline.from_pretrained(
            "kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16
        )
-        pipeline = pipeline.enable_model_cpu_offload()
+        pipeline.enable_model_cpu_offload()
        pipeline.set_progress_bar_config(disable=None)

        prompt = "red cat, 4k photo"

-        generator = torch.Generator(device="cuda").manual_seed(0)
+        generator = torch.Generator(device="cpu").manual_seed(0)
        image_emb, zero_image_emb = pipe_prior(
            prompt,
            generator=generator,
@@ -261,7 +262,7 @@ class KandinskyV22PipelineIntegrationTests(unittest.TestCase):
            negative_prompt="",
        ).to_tuple()

-        generator = torch.Generator(device="cuda").manual_seed(0)
+        generator = torch.Generator(device="cpu").manual_seed(0)
        output = pipeline(
            image_embeds=image_emb,
            negative_image_embeds=zero_image_emb,
@@ -269,9 +270,8 @@ class KandinskyV22PipelineIntegrationTests(unittest.TestCase):
            num_inference_steps=3,
            output_type="np",
        )
-
        image = output.images[0]
-
        assert image.shape == (512, 512, 3)

-        assert_mean_pixel_difference(image, expected_image)
+        max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
+        assert max_diff < 1e-4
--- a/tests/pipelines/kandinsky2_2/test_kandinsky_controlnet.py
+++ b/tests/pipelines/kandinsky2_2/test_kandinsky_controlnet.py
@@ -33,10 +33,11 @@ from diffusers.utils.testing_utils import (
    load_image,
    load_numpy,
    nightly,
+    numpy_cosine_similarity_distance,
    require_torch_gpu,
 )

-from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
+from ..test_pipelines_common import PipelineTesterMixin


 enable_full_determinism()
@@ -260,12 +261,12 @@ class KandinskyV22ControlnetPipelineIntegrationTests(unittest.TestCase):
        pipeline = KandinskyV22ControlnetPipeline.from_pretrained(
            "kandinsky-community/kandinsky-2-2-controlnet-depth", torch_dtype=torch.float16
        )
-        pipeline = pipeline.enable_model_cpu_offload()
+        pipeline.enable_model_cpu_offload()
        pipeline.set_progress_bar_config(disable=None)

        prompt = "A robot, 4k photo"

-        generator = torch.Generator(device="cuda").manual_seed(0)
+        generator = torch.Generator(device="cpu").manual_seed(0)
        image_emb, zero_image_emb = pipe_prior(
            prompt,
            generator=generator,
@@ -273,7 +274,7 @@ class KandinskyV22ControlnetPipelineIntegrationTests(unittest.TestCase):
            negative_prompt="",
        ).to_tuple()

-        generator = torch.Generator(device="cuda").manual_seed(0)
+        generator = torch.Generator(device="cpu").manual_seed(0)
        output = pipeline(
            image_embeds=image_emb,
            negative_image_embeds=zero_image_emb,
@@ -287,4 +288,5 @@ class KandinskyV22ControlnetPipelineIntegrationTests(unittest.TestCase):

        assert image.shape == (512, 512, 3)

-        assert_mean_pixel_difference(image, expected_image)
+        max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
+        assert max_diff < 1e-4
--- a/tests/pipelines/kandinsky2_2/test_kandinsky_controlnet_img2img.py
+++ b/tests/pipelines/kandinsky2_2/test_kandinsky_controlnet_img2img.py
@@ -34,10 +34,11 @@ from diffusers.utils.testing_utils import (
    load_image,
    load_numpy,
    nightly,
+    numpy_cosine_similarity_distance,
    require_torch_gpu,
 )

-from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
+from ..test_pipelines_common import PipelineTesterMixin


 enable_full_determinism()
@@ -274,7 +275,7 @@ class KandinskyV22ControlnetImg2ImgPipelineIntegrationTests(unittest.TestCase):
        pipeline = KandinskyV22ControlnetImg2ImgPipeline.from_pretrained(
            "kandinsky-community/kandinsky-2-2-controlnet-depth", torch_dtype=torch.float16
        )
-        pipeline = pipeline.enable_model_cpu_offload()
+        pipeline.enable_model_cpu_offload()

        pipeline.set_progress_bar_config(disable=None)

@@ -289,6 +290,7 @@ class KandinskyV22ControlnetImg2ImgPipelineIntegrationTests(unittest.TestCase):
            num_inference_steps=5,
        ).to_tuple()

+        generator = torch.Generator(device="cpu").manual_seed(0)
        output = pipeline(
            image=init_image,
            image_embeds=image_emb,
@@ -306,4 +308,5 @@ class KandinskyV22ControlnetImg2ImgPipelineIntegrationTests(unittest.TestCase):

        assert image.shape == (512, 512, 3)

-        assert_mean_pixel_difference(image, expected_image)
+        max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
+        assert max_diff < 1e-4
--- a/tests/pipelines/kandinsky2_2/test_kandinsky_img2img.py
+++ b/tests/pipelines/kandinsky2_2/test_kandinsky_img2img.py
@@ -33,11 +33,12 @@ from diffusers.utils.testing_utils import (
    floats_tensor,
    load_image,
    load_numpy,
+    numpy_cosine_similarity_distance,
    require_torch_gpu,
    slow,
 )

-from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
+from ..test_pipelines_common import PipelineTesterMixin


 enable_full_determinism()
@@ -270,8 +271,7 @@ class KandinskyV22Img2ImgPipelineIntegrationTests(unittest.TestCase):
        pipeline = KandinskyV22Img2ImgPipeline.from_pretrained(
            "kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16
        )
-        pipeline = pipeline.enable_model_cpu_offload()
-
+        pipeline.enable_model_cpu_offload()
        pipeline.set_progress_bar_config(disable=None)

        generator = torch.Generator(device="cpu").manual_seed(0)
@@ -282,6 +282,7 @@ class KandinskyV22Img2ImgPipelineIntegrationTests(unittest.TestCase):
            negative_prompt="",
        ).to_tuple()

+        generator = torch.Generator(device="cpu").manual_seed(0)
        output = pipeline(
            image=init_image,
            image_embeds=image_emb,
@@ -298,4 +299,5 @@ class KandinskyV22Img2ImgPipelineIntegrationTests(unittest.TestCase):

        assert image.shape == (768, 768, 3)

-        assert_mean_pixel_difference(image, expected_image)
+        max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
+        assert max_diff < 1e-4
--- a/tests/pipelines/kandinsky2_2/test_kandinsky_inpaint.py
+++ b/tests/pipelines/kandinsky2_2/test_kandinsky_inpaint.py
@@ -34,12 +34,13 @@ from diffusers.utils.testing_utils import (
    is_flaky,
    load_image,
    load_numpy,
+    numpy_cosine_similarity_distance,
    require_torch_gpu,
    slow,
    torch_device,
 )

-from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
+from ..test_pipelines_common import PipelineTesterMixin


 enable_full_determinism()
@@ -338,6 +339,7 @@ class KandinskyV22InpaintPipelineIntegrationTests(unittest.TestCase):
            negative_prompt="",
        ).to_tuple()

+        generator = torch.Generator(device="cpu").manual_seed(0)
        output = pipeline(
            image=init_image,
            mask_image=mask,
@@ -354,4 +356,5 @@ class KandinskyV22InpaintPipelineIntegrationTests(unittest.TestCase):

        assert image.shape == (768, 768, 3)

-        assert_mean_pixel_difference(image, expected_image)
+        max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
+        assert max_diff < 1e-4
--- a/tests/pipelines/pixart_alpha/test_pixart.py
+++ b/tests/pipelines/pixart_alpha/test_pixart.py
@@ -324,6 +324,10 @@ class PixArtAlphaPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
    def test_inference_batch_single_identical(self):
        self._test_inference_batch_single_identical(expected_max_diff=1e-3)

+    # PixArt transformer model does not work with sequential offload so skip it for now
+    def test_sequential_offload_forward_pass_twice(self):
+        pass
+

@slow
@require_torch_gpu
--- a/tests/pipelines/pixart_sigma/init.py
+++ b/tests/pipelines/pixart_sigma/init.py
--- a/tests/pipelines/pixart_sigma/test_pixart.py
+++ b/tests/pipelines/pixart_sigma/test_pixart.py
@@ -0,0 +1,430 @@
+# 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 gc
+import tempfile
+import unittest
+
+import numpy as np
+import torch
+from transformers import AutoTokenizer, T5EncoderModel
+
+from diffusers import (
+    AutoencoderKL,
+    DDIMScheduler,
+    PixArtSigmaPipeline,
+    Transformer2DModel,
+)
+from diffusers.utils.testing_utils import (
+    enable_full_determinism,
+    numpy_cosine_similarity_distance,
+    require_torch_gpu,
+    slow,
+    torch_device,
+)
+
+from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
+from ..test_pipelines_common import PipelineTesterMixin, to_np
+
+
+enable_full_determinism()
+
+
+class PixArtSigmaPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = PixArtSigmaPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+
+    required_optional_params = PipelineTesterMixin.required_optional_params
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        transformer = Transformer2DModel(
+            sample_size=8,
+            num_layers=2,
+            patch_size=2,
+            attention_head_dim=8,
+            num_attention_heads=3,
+            caption_channels=32,
+            in_channels=4,
+            cross_attention_dim=24,
+            out_channels=8,
+            attention_bias=True,
+            activation_fn="gelu-approximate",
+            num_embeds_ada_norm=1000,
+            norm_type="ada_norm_single",
+            norm_elementwise_affine=False,
+            norm_eps=1e-6,
+        )
+        torch.manual_seed(0)
+        vae = AutoencoderKL()
+
+        scheduler = DDIMScheduler()
+        text_encoder = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        components = {
+            "transformer": transformer.eval(),
+            "vae": vae.eval(),
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        inputs = {
+            "prompt": "A painting of a squirrel eating a burger",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 5.0,
+            "use_resolution_binning": False,
+            "output_type": "np",
+        }
+        return inputs
+
+    def test_sequential_cpu_offload_forward_pass(self):
+        # TODO(PVP, Sayak) need to fix later
+        return
+
+    def test_save_load_optional_components(self):
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(torch_device)
+
+        prompt = inputs["prompt"]
+        generator = inputs["generator"]
+        num_inference_steps = inputs["num_inference_steps"]
+        output_type = inputs["output_type"]
+
+        (
+            prompt_embeds,
+            prompt_attention_mask,
+            negative_prompt_embeds,
+            negative_prompt_attention_mask,
+        ) = pipe.encode_prompt(prompt)
+
+        # inputs with prompt converted to embeddings
+        inputs = {
+            "prompt_embeds": prompt_embeds,
+            "prompt_attention_mask": prompt_attention_mask,
+            "negative_prompt": None,
+            "negative_prompt_embeds": negative_prompt_embeds,
+            "negative_prompt_attention_mask": negative_prompt_attention_mask,
+            "generator": generator,
+            "num_inference_steps": num_inference_steps,
+            "output_type": output_type,
+            "use_resolution_binning": False,
+        }
+
+        # set all optional components to None
+        for optional_component in pipe._optional_components:
+            setattr(pipe, optional_component, None)
+
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        for optional_component in pipe._optional_components:
+            self.assertTrue(
+                getattr(pipe_loaded, optional_component) is None,
+                f"`{optional_component}` did not stay set to None after loading.",
+            )
+
+        inputs = self.get_dummy_inputs(torch_device)
+
+        generator = inputs["generator"]
+        num_inference_steps = inputs["num_inference_steps"]
+        output_type = inputs["output_type"]
+
+        # inputs with prompt converted to embeddings
+        inputs = {
+            "prompt_embeds": prompt_embeds,
+            "prompt_attention_mask": prompt_attention_mask,
+            "negative_prompt": None,
+            "negative_prompt_embeds": negative_prompt_embeds,
+            "negative_prompt_attention_mask": negative_prompt_attention_mask,
+            "generator": generator,
+            "num_inference_steps": num_inference_steps,
+            "output_type": output_type,
+            "use_resolution_binning": False,
+        }
+
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(to_np(output) - to_np(output_loaded)).max()
+        self.assertLess(max_diff, 1e-4)
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = pipe(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        self.assertEqual(image.shape, (1, 8, 8, 3))
+        expected_slice = np.array([0.6319, 0.3526, 0.3806, 0.6327, 0.4639, 0.4830, 0.2583, 0.5331, 0.4852])
+        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
+        self.assertLessEqual(max_diff, 1e-3)
+
+    def test_inference_non_square_images(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = pipe(**inputs, height=32, width=48).images
+        image_slice = image[0, -3:, -3:, -1]
+        self.assertEqual(image.shape, (1, 32, 48, 3))
+
+        expected_slice = np.array([0.6493, 0.5370, 0.4081, 0.4762, 0.3695, 0.4711, 0.3026, 0.5218, 0.5263])
+        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
+        self.assertLessEqual(max_diff, 1e-3)
+
+    def test_inference_with_embeddings_and_multiple_images(self):
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(torch_device)
+
+        prompt = inputs["prompt"]
+        generator = inputs["generator"]
+        num_inference_steps = inputs["num_inference_steps"]
+        output_type = inputs["output_type"]
+
+        prompt_embeds, prompt_attn_mask, negative_prompt_embeds, neg_prompt_attn_mask = pipe.encode_prompt(prompt)
+
+        # inputs with prompt converted to embeddings
+        inputs = {
+            "prompt_embeds": prompt_embeds,
+            "prompt_attention_mask": prompt_attn_mask,
+            "negative_prompt": None,
+            "negative_prompt_embeds": negative_prompt_embeds,
+            "negative_prompt_attention_mask": neg_prompt_attn_mask,
+            "generator": generator,
+            "num_inference_steps": num_inference_steps,
+            "output_type": output_type,
+            "num_images_per_prompt": 2,
+            "use_resolution_binning": False,
+        }
+
+        # set all optional components to None
+        for optional_component in pipe._optional_components:
+            setattr(pipe, optional_component, None)
+
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        for optional_component in pipe._optional_components:
+            self.assertTrue(
+                getattr(pipe_loaded, optional_component) is None,
+                f"`{optional_component}` did not stay set to None after loading.",
+            )
+
+        inputs = self.get_dummy_inputs(torch_device)
+
+        generator = inputs["generator"]
+        num_inference_steps = inputs["num_inference_steps"]
+        output_type = inputs["output_type"]
+
+        # inputs with prompt converted to embeddings
+        inputs = {
+            "prompt_embeds": prompt_embeds,
+            "prompt_attention_mask": prompt_attn_mask,
+            "negative_prompt": None,
+            "negative_prompt_embeds": negative_prompt_embeds,
+            "negative_prompt_attention_mask": neg_prompt_attn_mask,
+            "generator": generator,
+            "num_inference_steps": num_inference_steps,
+            "output_type": output_type,
+            "num_images_per_prompt": 2,
+            "use_resolution_binning": False,
+        }
+
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(to_np(output) - to_np(output_loaded)).max()
+        self.assertLess(max_diff, 1e-4)
+
+    def test_inference_with_multiple_images_per_prompt(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        inputs["num_images_per_prompt"] = 2
+        image = pipe(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        self.assertEqual(image.shape, (2, 8, 8, 3))
+        expected_slice = np.array([0.6319, 0.3526, 0.3806, 0.6327, 0.4639, 0.4830, 0.2583, 0.5331, 0.4852])
+        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
+        self.assertLessEqual(max_diff, 1e-3)
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(expected_max_diff=1e-3)
+
+    # PixArt transformer model does not work with sequential offload so skip it for now
+    def test_sequential_offload_forward_pass_twice(self):
+        pass
+
+
+@slow
+@require_torch_gpu
+class PixArtSigmaPipelineIntegrationTests(unittest.TestCase):
+    ckpt_id_1024 = "PixArt-alpha/PixArt-Sigma-XL-2-1024-MS"
+    ckpt_id_512 = "PixArt-alpha/PixArt-Sigma-XL-2-512-MS"
+    prompt = "A small cactus with a happy face in the Sahara desert."
+
+    def setUp(self):
+        super().setUp()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def tearDown(self):
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_pixart_1024(self):
+        generator = torch.Generator("cpu").manual_seed(0)
+
+        pipe = PixArtSigmaPipeline.from_pretrained(self.ckpt_id_1024, torch_dtype=torch.float16)
+        pipe.enable_model_cpu_offload()
+        prompt = self.prompt
+
+        image = pipe(prompt, generator=generator, num_inference_steps=2, output_type="np").images
+
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = np.array([0.0742, 0.0835, 0.2114, 0.0295, 0.0784, 0.2361, 0.1738, 0.2251, 0.3589])
+
+        max_diff = numpy_cosine_similarity_distance(image_slice.flatten(), expected_slice)
+        self.assertLessEqual(max_diff, 1e-4)
+
+    def test_pixart_512(self):
+        generator = torch.Generator("cpu").manual_seed(0)
+
+        pipe = PixArtSigmaPipeline.from_pretrained(self.ckpt_id_512, torch_dtype=torch.float16)
+        pipe.enable_model_cpu_offload()
+
+        prompt = self.prompt
+
+        image = pipe(prompt, generator=generator, num_inference_steps=2, output_type="np").images
+
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = np.array([0.3477, 0.3882, 0.4541, 0.3413, 0.3821, 0.4463, 0.4001, 0.4409, 0.4958])
+
+        max_diff = numpy_cosine_similarity_distance(image_slice.flatten(), expected_slice)
+        self.assertLessEqual(max_diff, 1e-4)
+
+    def test_pixart_1024_without_resolution_binning(self):
+        generator = torch.manual_seed(0)
+
+        pipe = PixArtSigmaPipeline.from_pretrained(self.ckpt_id_1024, torch_dtype=torch.float16)
+        pipe.enable_model_cpu_offload()
+
+        prompt = self.prompt
+        height, width = 1024, 768
+        num_inference_steps = 2
+
+        image = pipe(
+            prompt,
+            height=height,
+            width=width,
+            generator=generator,
+            num_inference_steps=num_inference_steps,
+            output_type="np",
+        ).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        generator = torch.manual_seed(0)
+        no_res_bin_image = pipe(
+            prompt,
+            height=height,
+            width=width,
+            generator=generator,
+            num_inference_steps=num_inference_steps,
+            output_type="np",
+            use_resolution_binning=False,
+        ).images
+        no_res_bin_image_slice = no_res_bin_image[0, -3:, -3:, -1]
+
+        assert not np.allclose(image_slice, no_res_bin_image_slice, atol=1e-4, rtol=1e-4)
+
+    def test_pixart_512_without_resolution_binning(self):
+        generator = torch.manual_seed(0)
+
+        pipe = PixArtSigmaPipeline.from_pretrained(self.ckpt_id_512, torch_dtype=torch.float16)
+        pipe.enable_model_cpu_offload()
+
+        prompt = self.prompt
+        height, width = 512, 768
+        num_inference_steps = 2
+
+        image = pipe(
+            prompt,
+            height=height,
+            width=width,
+            generator=generator,
+            num_inference_steps=num_inference_steps,
+            output_type="np",
+        ).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        generator = torch.manual_seed(0)
+        no_res_bin_image = pipe(
+            prompt,
+            height=height,
+            width=width,
+            generator=generator,
+            num_inference_steps=num_inference_steps,
+            output_type="np",
+            use_resolution_binning=False,
+        ).images
+        no_res_bin_image_slice = no_res_bin_image[0, -3:, -3:, -1]
+
+        assert not np.allclose(image_slice, no_res_bin_image_slice, atol=1e-4, rtol=1e-4)
--- a/tests/pipelines/test_pipelines_common.py
+++ b/tests/pipelines/test_pipelines_common.py
@@ -120,7 +120,7 @@ class SDFunctionTesterMixin:
        inputs["return_dict"] = False
        output_2 = pipe(**inputs)[0]

-        assert np.abs(output_2 - output_1).max() < 5e-1
+        assert np.abs(to_np(output_2) - to_np(output_1)).max() < 5e-1

        # test that tiled decode works with various shapes
        shapes = [(1, 4, 73, 97), (1, 4, 97, 73), (1, 4, 49, 65), (1, 4, 65, 49)]
--- a/tests/pipelines/wuerstchen/test_wuerstchen_prior.py
+++ b/tests/pipelines/wuerstchen/test_wuerstchen_prior.py
@@ -194,7 +194,7 @@ class WuerstchenPriorPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
    @skip_mps
    def test_inference_batch_single_identical(self):
        self._test_inference_batch_single_identical(
-            expected_max_diff=2e-1,
+            expected_max_diff=3e-1,
        )

    @skip_mps
--- a/utils/update_metadata.py
+++ b/utils/update_metadata.py
@@ -30,7 +30,7 @@ import tempfile

 import pandas as pd
 from datasets import Dataset
-from huggingface_hub import upload_folder
+from huggingface_hub import hf_hub_download, upload_folder

 from diffusers.pipelines.auto_pipeline import (
    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
@@ -39,6 +39,9 @@ from diffusers.pipelines.auto_pipeline import (
 )


+PIPELINE_TAG_JSON = "pipeline_tags.json"
+
+
 def get_supported_pipeline_table() -> dict:
    """
    Generates a dictionary containing the supported auto classes for each pipeline type,
@@ -57,8 +60,8 @@ def get_supported_pipeline_table() -> dict:
        (class_name.__name__, "image-to-image", "AutoPipelineForInpainting")
        for _, class_name in AUTO_INPAINT_PIPELINES_MAPPING.items()
    ]
-    all_supported_pipeline_classes.sort(key=lambda x: x[0])
    all_supported_pipeline_classes = list(set(all_supported_pipeline_classes))
+    all_supported_pipeline_classes.sort(key=lambda x: x[0])

    data = {}
    data["pipeline_class"] = [sample[0] for sample in all_supported_pipeline_classes]
@@ -79,8 +82,24 @@ def update_metadata(commit_sha: str):
    pipelines_table = pd.DataFrame(pipelines_table)
    pipelines_dataset = Dataset.from_pandas(pipelines_table)

+    hub_pipeline_tags_json = hf_hub_download(
+        repo_id="huggingface/diffusers-metadata",
+        filename=PIPELINE_TAG_JSON,
+        repo_type="dataset",
+    )
+    with open(hub_pipeline_tags_json) as f:
+        hub_pipeline_tags_json = f.read()
+
    with tempfile.TemporaryDirectory() as tmp_dir:
-        pipelines_dataset.to_json(os.path.join(tmp_dir, "pipeline_tags.json"))
+        pipelines_dataset.to_json(os.path.join(tmp_dir, PIPELINE_TAG_JSON))
+
+        with open(os.path.join(tmp_dir, PIPELINE_TAG_JSON)) as f:
+            pipeline_tags_json = f.read()
+
+        hub_pipeline_tags_equal = hub_pipeline_tags_json == pipeline_tags_json
+        if hub_pipeline_tags_equal:
+            print("No updates, not pushing the metadata files.")
+            return

        if commit_sha is not None:
            commit_message = (
Author	SHA1	Message	Date
sayakpaul	b0b58d0fc6	decorate UNetControlNetXSModelTests::test_forward_no_control with is_flaky	2024-04-25 07:03:42 +05:30
Sayak Paul	e963621649	[PixArt] fix small nits in pixart sigma (#7767 ) fix small nits in pixart sigma	2024-04-25 06:37:35 +05:30
Junsong Chen	39215aa30e	PixArt-Sigma Implementation (#7654 ) * support PixArt-DMD --------- Co-authored-by: jschen <chenjunsong4@h-partners.com> Co-authored-by: badayvedat <badayvedat@gmail.com> Co-authored-by: Vedat Baday <54285744+badayvedat@users.noreply.github.com> Co-authored-by: Sayak Paul <spsayakpaul@gmail.com> Co-authored-by: YiYi Xu <yixu310@gmail.com> Co-authored-by: yiyixuxu <yixu310@gmail,com>	2024-04-23 22:33:08 -10:00
Dhruv Nair	9ef43f38d4	Fix test for consistency decoder. (#7746 ) update	2024-04-24 12:28:11 +05:30
Dhruv Nair	88018fcf20	Fix failing VAE tiling test (#7747 ) update	2024-04-24 12:27:45 +05:30
Steven Liu	7404f1e9dc	[docs] Clean up toctree (#7715 ) * toctree * optim * feedback * improve overview	2024-04-23 09:30:33 -07:00
Sayak Paul	5a69227863	[Metadat utils] fix: json lines ordering. (#7744 ) fix: json lines ordering.	2024-04-23 14:32:30 +05:30
Sai-Suraj-27	fc9fecc217	fix: Fixed a wrong decorator by modifying it to `@classmethod` (#7653 ) * Fixed wrong decorator by modifying it to @classmethod. * Updated the method and it's argument. --------- Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>	2024-04-22 14:41:35 -10:00
Fabio Rigano	065f251766	Restore AttnProcessor2_0 in unload_ip_adapter (#7727 ) * Restore AttnProcessor2_0 in unload_ip_adapter * Fix style * Update test --------- Co-authored-by: YiYi Xu <yixu310@gmail.com>	2024-04-22 13:59:03 -10:00
Jenyuan-Huang	21c747fa0f	Support InstantStyle (#7668 ) * enable control ip-adapter per-transformer block on-the-fly --------- Co-authored-by: sayakpaul <spsayakpaul@gmail.com> Co-authored-by: ResearcherXman <xhs.research@gmail.com> Co-authored-by: YiYi Xu <yixu310@gmail.com>	2024-04-22 13:20:19 -10:00
Phil Butler	09129842e7	Remove redundant lines (#7396 ) Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com> Co-authored-by: YiYi Xu <yixu310@gmail.com>	2024-04-22 09:32:16 -10:00
Steven Liu	33b363edfa	[docs] AutoPipeline (#7714 ) * autopipeline * edits * feedback --------- Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>	2024-04-22 10:15:07 -07:00
Dhruv Nair	a9dd86029e	Fix Kandinksy V22 tests (#7699 ) update	2024-04-22 15:41:59 +05:30
Dhruv Nair	9100652494	Update Wuerschten Test (#7700 ) update	2024-04-22 15:41:39 +05:30