AnimateDiff Video to Video (#6328)

* begin animatediff img2video and video2video

* revert animatediff to original implementation

* add img2video as pipeline

* update

* add vid2vid pipeline

* update imports

* update

* remove copied from line for check_inputs

* update

* update examples

* add multi-batch support

* fix __init__.py files

* move img2vid to community

* update community readme and examples

* fix

* make fix-copies

* add vid2vid batch params

* apply suggestions from review

Co-Authored-By: Dhruv Nair <[email protected]>

* add test for animatediff vid2vid

* torch.stack -> torch.cat

Co-Authored-By: Dhruv Nair <[email protected]>

* make style

* docs for vid2vid

* update

* fix prepare_latents

* fix docs

* remove img2vid

* update README to :main

* remove slow test

* refactor pipeline output

* update docs

* update docs

* merge community readme from :main

* final fix i promise

* add support for url in animatediff example

* update example

* update callbacks to latest implementation

* Update src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py

Co-authored-by: Patrick von Platen <[email protected]>

* Update src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py

Co-authored-by: Patrick von Platen <[email protected]>

* fix merge

* Apply suggestions from code review

* remove callback and callback_steps as suggested in review

* Update tests/pipelines/animatediff/test_animatediff_video2video.py

Co-authored-by: Patrick von Platen <[email protected]>

* fix import error caused due to unet refactor in #6630

* fix numpy import error after tensor2vid refactor in #6626

* make fix-copies

* fix numpy error

* fix progress bar test

---------

Co-authored-by: Dhruv Nair <[email protected]>
Co-authored-by: Patrick von Platen <[email protected]>

Author: 3 people

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

@@ -25,13 +25,16 @@ The abstract of the paper is the following:
 | Pipeline | Tasks | Demo
 |---|---|:---:|
 | [AnimateDiffPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/animatediff/pipeline_animatediff.py) | *Text-to-Video Generation with AnimateDiff* |
+| [AnimateDiffVideoToVideoPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py) | *Video-to-Video Generation with AnimateDiff* |
 
 ## Available checkpoints
 
 Motion Adapter checkpoints can be found under [guoyww](https://huggingface.co/guoyww/). These checkpoints are meant to work with any model based on Stable Diffusion 1.4/1.5.
 
 ## Usage example
 
+### AnimateDiffPipeline
+
 AnimateDiff works with a MotionAdapter checkpoint and a Stable Diffusion model checkpoint. The MotionAdapter is a collection of Motion Modules that are responsible for adding coherent motion across image frames. These modules are applied after the Resnet and Attention blocks in Stable Diffusion UNet.
 
 The following example demonstrates how to use a *MotionAdapter* checkpoint with Diffusers for inference based on StableDiffusion-1.4/1.5.
@@ -98,6 +101,114 @@ AnimateDiff tends to work better with finetuned Stable Diffusion models. If you
 
 </Tip>
 
+### AnimateDiffVideoToVideoPipeline
+
+AnimateDiff can also be used to generate visually similar videos or enable style/character/background or other edits starting from an initial video, allowing you to seamlessly explore creative possibilities.
+
+```python
+import imageio
+import requests
+import torch
+from diffusers import AnimateDiffVideoToVideoPipeline, DDIMScheduler, MotionAdapter
+from diffusers.utils import export_to_gif
+from io import BytesIO
+from PIL import Image
+
+# Load the motion adapter
+adapter = MotionAdapter.from_pretrained("guoyww/animatediff-motion-adapter-v1-5-2", torch_dtype=torch.float16)
+# load SD 1.5 based finetuned model
+model_id = "SG161222/Realistic_Vision_V5.1_noVAE"
+pipe = AnimateDiffVideoToVideoPipeline.from_pretrained(model_id, motion_adapter=adapter, torch_dtype=torch.float16)
+scheduler = DDIMScheduler.from_pretrained(
+    model_id,
+    subfolder="scheduler",
+    clip_sample=False,
+    timestep_spacing="linspace",
+    beta_schedule="linear",
+    steps_offset=1,
+)
+pipe.scheduler = scheduler
+
+# enable memory savings
+pipe.enable_vae_slicing()
+pipe.enable_model_cpu_offload()
+
+# helper function to load videos
+def load_video(file_path: str):
+    images = []
+
+    if file_path.startswith(('http://', 'https://')):
+        # If the file_path is a URL
+        response = requests.get(file_path)
+        response.raise_for_status()
+        content = BytesIO(response.content)
+        vid = imageio.get_reader(content)
+    else:
+        # Assuming it's a local file path
+        vid = imageio.get_reader(file_path)
+
+    for frame in vid:
+        pil_image = Image.fromarray(frame)
+        images.append(pil_image)
+
+    return images
+
+video = load_video("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff-vid2vid-input-1.gif")
+
+output = pipe(
+    video = video,
+    prompt="panda playing a guitar, on a boat, in the ocean, high quality",
+    negative_prompt="bad quality, worse quality",
+    guidance_scale=7.5,
+    num_inference_steps=25,
+    strength=0.5,
+    generator=torch.Generator("cpu").manual_seed(42),
+)
+frames = output.frames[0]
+export_to_gif(frames, "animation.gif")
+```
+
+Here are some sample outputs:
+
+<table>
+    <tr>
+      <th align=center>Source Video</th>
+      <th align=center>Output Video</th>
+    </tr>
+    <tr>
+        <td align=center>
+          raccoon playing a guitar
+          <br />
+          <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff-vid2vid-input-1.gif"
+              alt="racoon playing a guitar"
+              style="width: 300px;" />
+        </td>
+        <td align=center>
+          panda playing a guitar
+          <br/>
+          <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff-vid2vid-output-1.gif"
+              alt="panda playing a guitar"
+              style="width: 300px;" />
+        </td>
+    </tr>
+    <tr>
+        <td align=center>
+          closeup of margot robbie, fireworks in the background, high quality
+          <br />
+          <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff-vid2vid-input-2.gif"
+              alt="closeup of margot robbie, fireworks in the background, high quality"
+              style="width: 300px;" />
+        </td>
+        <td align=center>
+          closeup of tony stark, robert downey jr, fireworks
+          <br/>
+          <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff-vid2vid-output-2.gif"
+              alt="closeup of tony stark, robert downey jr, fireworks"
+              style="width: 300px;" />
+        </td>
+    </tr>
+</table>
+
 ## Using Motion LoRAs
 
 Motion LoRAs are a collection of LoRAs that work with the `guoyww/animatediff-motion-adapter-v1-5-2` checkpoint. These LoRAs are responsible for adding specific types of motion to the animations.

src/diffusers/__init__.py

@@ -208,6 +208,7 @@
             "AmusedInpaintPipeline",
             "AmusedPipeline",
             "AnimateDiffPipeline",
+            "AnimateDiffVideoToVideoPipeline",
             "AudioLDM2Pipeline",
             "AudioLDM2ProjectionModel",
             "AudioLDM2UNet2DConditionModel",
@@ -569,6 +570,7 @@
             AmusedInpaintPipeline,
             AmusedPipeline,
             AnimateDiffPipeline,
+            AnimateDiffVideoToVideoPipeline,
             AudioLDM2Pipeline,
             AudioLDM2ProjectionModel,
             AudioLDM2UNet2DConditionModel,

src/diffusers/pipelines/__init__.py

@@ -109,7 +109,10 @@
         ]
     )
     _import_structure["amused"] = ["AmusedImg2ImgPipeline", "AmusedInpaintPipeline", "AmusedPipeline"]
-    _import_structure["animatediff"] = ["AnimateDiffPipeline"]
+    _import_structure["animatediff"] = [
+        "AnimateDiffPipeline",
+        "AnimateDiffVideoToVideoPipeline",
+    ]
     _import_structure["audioldm"] = ["AudioLDMPipeline"]
     _import_structure["audioldm2"] = [
         "AudioLDM2Pipeline",
@@ -341,7 +344,7 @@
         from ..utils.dummy_torch_and_transformers_objects import *
     else:
         from .amused import AmusedImg2ImgPipeline, AmusedInpaintPipeline, AmusedPipeline
-        from .animatediff import AnimateDiffPipeline
+        from .animatediff import AnimateDiffPipeline, AnimateDiffVideoToVideoPipeline
         from .audioldm import AudioLDMPipeline
         from .audioldm2 import (
             AudioLDM2Pipeline,

src/diffusers/pipelines/animatediff/__init__.py

@@ -11,7 +11,7 @@
 
 
 _dummy_objects = {}
-_import_structure = {}
+_import_structure = {"pipeline_output": ["AnimateDiffPipelineOutput"]}
 
 try:
     if not (is_transformers_available() and is_torch_available()):
@@ -21,7 +21,8 @@
 
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
-    _import_structure["pipeline_animatediff"] = ["AnimateDiffPipeline", "AnimateDiffPipelineOutput"]
+    _import_structure["pipeline_animatediff"] = ["AnimateDiffPipeline"]
+    _import_structure["pipeline_animatediff_video2video"] = ["AnimateDiffVideoToVideoPipeline"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
@@ -31,7 +32,9 @@
         from ...utils.dummy_torch_and_transformers_objects import *
 
     else:
-        from .pipeline_animatediff import AnimateDiffPipeline, AnimateDiffPipelineOutput
+        from .pipeline_animatediff import AnimateDiffPipeline
+        from .pipeline_animatediff_video2video import AnimateDiffVideoToVideoPipeline
+        from .pipeline_output import AnimateDiffPipelineOutput
 
 else:
     import sys

src/diffusers/pipelines/animatediff/pipeline_animatediff.py

@@ -14,7 +14,6 @@
 
 import inspect
 import math
-from dataclasses import dataclass
 from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
 import numpy as np
@@ -37,7 +36,6 @@
 )
 from ...utils import (
     USE_PEFT_BACKEND,
-    BaseOutput,
     deprecate,
     logging,
     replace_example_docstring,
@@ -46,6 +44,7 @@
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import AnimateDiffPipelineOutput
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -153,11 +152,6 @@ def _freq_mix_3d(x: torch.Tensor, noise: torch.Tensor, LPF: torch.Tensor) -> tor
     return x_mixed
 
 
-@dataclass
-class AnimateDiffPipelineOutput(BaseOutput):
-    frames: Union[torch.Tensor, np.ndarray]
-
-
 class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdapterMixin, LoraLoaderMixin):
     r"""
     Pipeline for text-to-video generation.