huggingface
diff --git a/‎PHILOSOPHY.md‎
Lines changed: 1 addition & 1 deletion b/‎PHILOSOPHY.md‎
Lines changed: 1 addition & 1 deletion
diff --git a/‎docs/source/en/_toctree.yml‎
Lines changed: 2 additions & 0 deletions b/‎docs/source/en/_toctree.yml‎
Lines changed: 2 additions & 0 deletions
diff --git a/‎docs/source/en/using-diffusers/svd.md‎
Lines changed: 131 additions & 0 deletions b/‎docs/source/en/using-diffusers/svd.md‎
Lines changed: 131 additions & 0 deletions
@@ -82,7 +82,7 @@ Models are designed as configurable toolboxes that are natural extensions of [Py
 The following design principles are followed:
 - Models correspond to **a type of model architecture**. *E.g.* the [`UNet2DConditionModel`] class is used for all UNet variations that expect 2D image inputs and are conditioned on some context.
 - All models can be found in [`src/diffusers/models`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models) and every model architecture shall be defined in its file, e.g. [`unet_2d_condition.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_condition.py), [`transformer_2d.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/transformer_2d.py), etc...
-- Models **do not** follow the single-file policy and should make use of smaller model building blocks, such as [`attention.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py), [`resnet.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py), [`embeddings.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/embeddings.py), etc... **Note**: This is in stark contrast to Transformers' modelling files and shows that models do not really follow the single-file policy.
+- Models **do not** follow the single-file policy and should make use of smaller model building blocks, such as [`attention.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py), [`resnet.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py), [`embeddings.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/embeddings.py), etc... **Note**: This is in stark contrast to Transformers' modeling files and shows that models do not really follow the single-file policy.
 - Models intend to expose complexity, just like PyTorch's `Module` class, and give clear error messages.
 - Models all inherit from `ModelMixin` and `ConfigMixin`.
 - Models can be optimized for performance when it doesn’t demand major code changes, keep backward compatibility, and give significant memory or compute gain.
 
@@ -94,6 +94,8 @@
       title: Latent Consistency Model-LoRA
     - local: using-diffusers/inference_with_lcm
       title: Latent Consistency Model
+    - local: using-diffusers/svd
+      title: Stable Video Diffusion
     title: Specific pipeline examples
   - sections:
     - local: training/overview
 
@@ -0,0 +1,131 @@
+<!--Copyright 2023 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.
+-->
+
+# Stable Video Diffusion
+
+[[open-in-colab]]
+
+[Stable Video Diffusion](https://static1.squarespace.com/static/6213c340453c3f502425776e/t/655ce779b9d47d342a93c890/1700587395994/stable_video_diffusion.pdf) is a powerful image-to-video generation model that can generate high resolution (576x1024) 2-4 second videos conditioned on the input image.
+
+This guide will show you how to use SVD to short generate videos from images.
+
+Before you begin, make sure you have the following libraries installed:
+
+```py
+!pip install -q -U diffusers transformers accelerate 
+```
+
+## Image to Video Generation
+
+The are two variants of SVD. [SVD](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid) 
+and [SVD-XT](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt). The svd checkpoint is trained to generate 14 frames and the svd-xt checkpoint is further 
+finetuned to generate 25 frames.
+
+We will use the `svd-xt` checkpoint for this guide.
+
+```python
+import torch
+
+from diffusers import StableVideoDiffusionPipeline
+from diffusers.utils import load_image, export_to_video
+
+pipe = StableVideoDiffusionPipeline.from_pretrained(
+    "stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16"
+)
+pipe.enable_model_cpu_offload()
+
+# Load the conditioning image
+image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png?download=true")
+image = image.resize((1024, 576))
+
+generator = torch.manual_seed(42)
+frames = pipe(image, decode_chunk_size=8, generator=generator).frames[0]
+
+export_to_video(frames, "generated.mp4", fps=7)
+```
+
+<video width="1024" height="576" controls>
+  <source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket_generated.mp4?download=true" type="video/mp4">
+</video>
+
+<Tip>
+Since generating videos is more memory intensive we can use the `decode_chunk_size` argument to control how many frames are decoded at once. This will reduce the memory usage. It's recommended to tweak this value based on your GPU memory.
+Setting `decode_chunk_size=1` will decode one frame at a time and will use the least amount of memory but the video might have some flickering.
+
+Additionally, we also use [model cpu offloading](../../optimization/memory#model-offloading) to reduce the memory usage.
+</Tip>
+
+
+### Torch.compile
+
+You can achieve a 20-25% speed-up at the expense of slightly increased memory by compiling the UNet as follows:
+
+```diff
+- pipe.enable_model_cpu_offload()
++ pipe.to("cuda")
++ pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+```
+
+### Low-memory
+
+Video generation is very memory intensive as we have to essentially generate `num_frames` all at once. The mechanism is very comparable to text-to-image generation with a high batch size. To reduce the memory requirement you have multiple options. The following options trade inference speed against lower memory requirement:
+- enable model offloading: Each component of the pipeline is offloaded to CPU once it's not needed anymore.
+- enable feed-forward chunking: The feed-forward layer runs in a loop instead of running with a single huge feed-forward batch size
+- reduce `decode_chunk_size`: This means that the VAE decodes frames in chunks instead of decoding them all together. **Note**: In addition to leading to a small slowdown, this method also slightly leads to video quality deterioration
+
+ You can enable them as follows:
+ ```diff
+-pipe.enable_model_cpu_offload()
+-frames = pipe(image, decode_chunk_size=8, generator=generator).frames[0]
++pipe.enable_model_cpu_offload()
++pipe.unet.enable_forward_chunking()
++frames = pipe(image, decode_chunk_size=2, generator=generator, num_frames=25).frames[0]
+```
+
+
+Including all these tricks should lower the memory requirement to less than 8GB VRAM.
+
+### Micro-conditioning
+
+Along with conditioning image Stable Diffusion Video also allows providing micro-conditioning that allows more control over the generated video.
+It accepts the following arguments:
+
+- `fps`: The frames per second of the generated video.
+- `motion_bucket_id`: The motion bucket id to use for the generated video. This can be used to control the motion of the generated video. Increasing the motion bucket id will increase the motion of the generated video.
+- `noise_aug_strength`: The amount of noise added to the conditioning image. The higher the values the less the video will resemble the conditioning image. Increasing this value will also increase the motion of the generated video.
+
+Here is an example of using micro-conditioning to generate a video with more motion.
+
+```python
+import torch
+
+from diffusers import StableVideoDiffusionPipeline
+from diffusers.utils import load_image, export_to_video
+
+pipe = StableVideoDiffusionPipeline.from_pretrained(
+    "stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16"
+)
+pipe.enable_model_cpu_offload()
+
+# Load the conditioning image
+image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png?download=true")
+image = image.resize((1024, 576))
+
+generator = torch.manual_seed(42)
+frames = pipe(image, decode_chunk_size=8, generator=generator, motion_bucket_id=180, noise_aug_strength=0.1).frames[0]
+export_to_video(frames, "generated.mp4", fps=7)
+```
+
+<video width="1024" height="576" controls>
+  <source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket_generated_motion.mp4?download=true" type="video/mp4">
+</video>
+