mirror of
https://github.com/Wan-Video/Wan2.1.git
synced 2025-06-05 14:54:54 +00:00
Merge a9d7494136 into b58b7c5737
This commit is contained in:
Bakhtiyor Sulaymonov
GitHub
commit
09f87601e2
2
.gitignore
vendored
2
.gitignore
vendored
@ -34,3 +34,5 @@ Wan2.1-T2V-14B/
|
||||
Wan2.1-T2V-1.3B/
|
||||
Wan2.1-I2V-14B-480P/
|
||||
Wan2.1-I2V-14B-720P/
|
||||
venv_wan/
|
||||
venv_wan_py310/
|
||||
|
||||
105
README.md
105
README.md
@ -1,31 +1,78 @@
|
||||
# Wan2.1
|
||||
# Wan2.1 Text-to-Video Model
|
||||
|
||||
<p align="center">
|
||||
<img src="assets/logo.png" width="400"/>
|
||||
<p>
|
||||
This repository contains the Wan2.1 text-to-video model, adapted for macOS with M1 Pro chip. This adaptation allows macOS users to run the model efficiently, overcoming CUDA-specific limitations.
|
||||
|
||||
<p align="center">
|
||||
💜 <a href=""><b>Wan</b></a>    ｜    🖥️ <a href="https://github.com/Wan-Video/Wan2.1">GitHub</a>    |   🤗 <a href="https://huggingface.co/Wan-AI/">Hugging Face</a>   |   🤖 <a href="https://modelscope.cn/organization/Wan-AI">ModelScope</a>   |    📑 <a href="">Paper (Coming soon)</a>    |    📑 <a href="https://wanxai.com">Blog</a>    |   💬 <a href="https://gw.alicdn.com/imgextra/i2/O1CN01tqjWFi1ByuyehkTSB_!!6000000000015-0-tps-611-1279.jpg">WeChat Group</a>   |    📖 <a href="https://discord.gg/AKNgpMK4Yj">Discord</a>  
|
||||
<br>
|
||||
|
||||
-----
|
||||
## Introduction
|
||||
The Wan2.1 model is an open-source text-to-video generation model. It transforms textual descriptions into video sequences, leveraging advanced machine learning techniques.
|
||||
|
||||
[**Wan: Open and Advanced Large-Scale Video Generative Models**]("") <be>
|
||||
## Changes for macOS
|
||||
|
||||
In this repository, we present **Wan2.1**, a comprehensive and open suite of video foundation models that pushes the boundaries of video generation. **Wan2.1** offers these key features:
|
||||
- 👍 **SOTA Performance**: **Wan2.1** consistently outperforms existing open-source models and state-of-the-art commercial solutions across multiple benchmarks.
|
||||
- 👍 **Supports Consumer-grade GPUs**: The T2V-1.3B model requires only 8.19 GB VRAM, making it compatible with almost all consumer-grade GPUs. It can generate a 5-second 480P video on an RTX 4090 in about 4 minutes (without optimization techniques like quantization). Its performance is even comparable to some closed-source models.
|
||||
- 👍 **Multiple Tasks**: **Wan2.1** excels in Text-to-Video, Image-to-Video, Video Editing, Text-to-Image, and Video-to-Audio, advancing the field of video generation.
|
||||
- 👍 **Visual Text Generation**: **Wan2.1** is the first video model capable of generating both Chinese and English text, featuring robust text generation that enhances its practical applications.
|
||||
- 👍 **Powerful Video VAE**: **Wan-VAE** delivers exceptional efficiency and performance, encoding and decoding 1080P videos of any length while preserving temporal information, making it an ideal foundation for video and image generation.
|
||||
This version includes modifications to make the model compatible with macOS, specifically for systems using the M1 Pro chip. Key changes include:
|
||||
|
||||
## Video Demos
|
||||
- Adaptation of CUDA-specific code to work with MPS (Metal Performance Shaders) on macOS.
|
||||
- Environment variable settings for MPS fallback to CPU for unsupported operations.
|
||||
- Adjustments to command-line arguments for better compatibility with macOS.
|
||||
|
||||
<div align="center">
|
||||
<video src="https://github.com/user-attachments/assets/4aca6063-60bf-4953-bfb7-e265053f49ef" width="70%" poster=""> </video>
|
||||
</div>
|
||||
## Installation Instructions
|
||||
|
||||
Follow these steps to set up the environment on macOS:
|
||||
|
||||
1. **Install Homebrew**: If not already installed, use Homebrew to manage packages.
|
||||
```bash
|
||||
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
|
||||
```
|
||||
|
||||
2. **Install Python 3.10+**:
|
||||
```bash
|
||||
brew install python@3.10
|
||||
```
|
||||
|
||||
3. **Create and Activate a Virtual Environment**:
|
||||
```bash
|
||||
python3.10 -m venv venv_wan
|
||||
source venv_wan/bin/activate
|
||||
```
|
||||
|
||||
4. **Install Dependencies**:
|
||||
```bash
|
||||
pip install -r requirements.txt
|
||||
pip install einops
|
||||
```
|
||||
|
||||
5. **Download models using huggingface-cli**:
|
||||
```bash
|
||||
pip install "huggingface_hub[cli]"
|
||||
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B --local-dir ./Wan2.1-T2V-1.3B
|
||||
```
|
||||
**Or download models using huggingface-cli**:
|
||||
```bash
|
||||
pip install modelscope
|
||||
modelscope download Wan-AI/Wan2.1-T2V-1.3B --local_dir ./Wan2.1-T2V-1.3B
|
||||
```
|
||||
|
||||
## Usage
|
||||
|
||||
To generate a video, use the following command:
|
||||
|
||||
```bash
|
||||
export PYTORCH_ENABLE_MPS_FALLBACK=1
|
||||
python generate.py --task t2v-1.3B --size "480*832" --frame_num 16 --sample_steps 25 --ckpt_dir ./Wan2.1-T2V-1.3B --offload_model True --t5_cpu --device mps --prompt "Lion running under snow in Samarkand" --save_file output_video.mp4
|
||||
```
|
||||
|
||||
## Optimization Tips
|
||||
|
||||
- **Use CPU for Large Models**: If you encounter memory issues, use `--device cpu`.
|
||||
- **Reduce Resolution and Frame Count**: Use smaller resolutions and fewer frames to reduce memory usage.
|
||||
- **Monitor System Resources**: Keep an eye on memory usage and adjust parameters as needed.
|
||||
|
||||
## Acknowledgments
|
||||
|
||||
This project is based on the original Wan2.1 model. Special thanks to the original authors and contributors for their work.
|
||||
|
||||
## 🔥 Latest News!!
|
||||
|
||||
* Mar 3, 2025: 👋 **Wan2.1**'s T2V and I2V have been integrated into Diffusers ([T2V](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan#diffusers.WanPipeline) | [I2V](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan#diffusers.WanImageToVideoPipeline)). Feel free to give it a try!
|
||||
* Feb 27, 2025: 👋 **Wan2.1** has been integrated into [ComfyUI](https://comfyanonymous.github.io/ComfyUI_examples/wan/). Enjoy!
|
||||
@ -36,6 +83,10 @@ If your work has improved **Wan2.1** and you would like more people to see it, p
|
||||
- [TeaCache](https://github.com/ali-vilab/TeaCache) now supports **Wan2.1** acceleration, capable of increasing speed by approximately 2x. Feel free to give it a try!
|
||||
- [DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) provides more support for **Wan2.1**, including video-to-video, FP8 quantization, VRAM optimization, LoRA training, and more. Please refer to [their examples](https://github.com/modelscope/DiffSynth-Studio/tree/main/examples/wanvideo).
|
||||
|
||||
<div align="center">
|
||||
<video src="https://github.com/user-attachments/assets/4aca6063-60bf-4953-bfb7-e265053f49ef" width="70%" poster=""> </video>
|
||||
</div>
|
||||
|
||||
|
||||
## 📑 Todo List
|
||||
- Wan2.1 Text-to-Video
|
||||
@ -49,6 +100,8 @@ If your work has improved **Wan2.1** and you would like more people to see it, p
|
||||
- [x] Multi-GPU Inference code of the 14B model
|
||||
- [x] Checkpoints of the 14B model
|
||||
- [x] Gradio demo
|
||||
- [X] ComfyUI integration
|
||||
- [ ] Diffusers integration
|
||||
- [x] ComfyUI integration
|
||||
- [x] Diffusers integration
|
||||
- [ ] Diffusers + Multi-GPU Inference
|
||||
@ -157,6 +210,14 @@ torchrun --nproc_per_node=8 generate.py --task t2v-14B --size 1280*720 --ckpt_di
|
||||
|
||||
Extending the prompts can effectively enrich the details in the generated videos, further enhancing the video quality. Therefore, we recommend enabling prompt extension. We provide the following two methods for prompt extension:
|
||||
|
||||
## Usage
|
||||
|
||||
To generate a video, use the following command:
|
||||
|
||||
```bash
|
||||
export PYTORCH_ENABLE_MPS_FALLBACK=1
|
||||
python generate.py --task t2v-1.3B --size "480*832" --frame_num 16 --sample_steps 25 --ckpt_dir ./Wan2.1-T2V-1.3B --offload_model True --t5_cpu --device mps --prompt "Lion running under snow in Samarkand" --save_file output_video.mp4
|
||||
```
|
||||
- Use the Dashscope API for extension.
|
||||
- Apply for a `dashscope.api_key` in advance ([EN](https://www.alibabacloud.com/help/en/model-studio/getting-started/first-api-call-to-qwen) | [CN](https://help.aliyun.com/zh/model-studio/getting-started/first-api-call-to-qwen)).
|
||||
- Configure the environment variable `DASH_API_KEY` to specify the Dashscope API key. For users of Alibaba Cloud's international site, you also need to set the environment variable `DASH_API_URL` to 'https://dashscope-intl.aliyuncs.com/api/v1'. For more detailed instructions, please refer to the [dashscope document](https://www.alibabacloud.com/help/en/model-studio/developer-reference/use-qwen-by-calling-api?spm=a2c63.p38356.0.i1).
|
||||
@ -485,7 +546,17 @@ The models in this repository are licensed under the Apache 2.0 License. We clai
|
||||
|
||||
We would like to thank the contributors to the [SD3](https://huggingface.co/stabilityai/stable-diffusion-3-medium), [Qwen](https://huggingface.co/Qwen), [umt5-xxl](https://huggingface.co/google/umt5-xxl), [diffusers](https://github.com/huggingface/diffusers) and [HuggingFace](https://huggingface.co) repositories, for their open research.
|
||||
|
||||
## Optimization Tips
|
||||
|
||||
- **Use CPU for Large Models**: If you encounter memory issues, use `--device cpu`.
|
||||
- **Reduce Resolution and Frame Count**: Use smaller resolutions and fewer frames to reduce memory usage.
|
||||
- **Monitor System Resources**: Keep an eye on memory usage and adjust parameters as needed.
|
||||
|
||||
## Acknowledgments
|
||||
|
||||
|
||||
This project is based on the original Wan2.1 model. Special thanks to the original authors and contributors for their work.
|
||||
|
||||
## Contact Us
|
||||
If you would like to leave a message to our research or product teams, feel free to join our [Discord](https://discord.gg/AKNgpMK4Yj) or [WeChat groups](https://gw.alicdn.com/imgextra/i2/O1CN01tqjWFi1ByuyehkTSB_!!6000000000015-0-tps-611-1279.jpg)!
|
||||
|
||||
|
||||
402
README_original.md
Normal file
402
README_original.md
Normal file
@ -0,0 +1,402 @@
|
||||
# Wan2.1
|
||||
|
||||
<p align="center">
|
||||
<img src="assets/logo.png" width="400"/>
|
||||
<p>
|
||||
|
||||
<p align="center">
|
||||
💜 <a href=""><b>Wan</b></a>    ｜    🖥️ <a href="https://github.com/Wan-Video/Wan2.1">GitHub</a>    |   🤗 <a href="https://huggingface.co/Wan-AI/">Hugging Face</a>   |   🤖 <a href="https://modelscope.cn/organization/Wan-AI">ModelScope</a>   |    📑 <a href="">Paper (Coming soon)</a>    |    📑 <a href="https://wanxai.com">Blog</a>    |   💬 <a href="https://gw.alicdn.com/imgextra/i2/O1CN01tqjWFi1ByuyehkTSB_!!6000000000015-0-tps-611-1279.jpg">WeChat Group</a>   |    📖 <a href="https://discord.gg/p5XbdQV7">Discord</a>  
|
||||
<br>
|
||||
|
||||
-----
|
||||
|
||||
[**Wan: Open and Advanced Large-Scale Video Generative Models**]("") <be>
|
||||
|
||||
In this repository, we present **Wan2.1**, a comprehensive and open suite of video foundation models that pushes the boundaries of video generation. **Wan2.1** offers these key features:
|
||||
- 👍 **SOTA Performance**: **Wan2.1** consistently outperforms existing open-source models and state-of-the-art commercial solutions across multiple benchmarks.
|
||||
- 👍 **Supports Consumer-grade GPUs**: The T2V-1.3B model requires only 8.19 GB VRAM, making it compatible with almost all consumer-grade GPUs. It can generate a 5-second 480P video on an RTX 4090 in about 4 minutes (without optimization techniques like quantization). Its performance is even comparable to some closed-source models.
|
||||
- 👍 **Multiple Tasks**: **Wan2.1** excels in Text-to-Video, Image-to-Video, Video Editing, Text-to-Image, and Video-to-Audio, advancing the field of video generation.
|
||||
- 👍 **Visual Text Generation**: **Wan2.1** is the first video model capable of generating both Chinese and English text, featuring robust text generation that enhances its practical applications.
|
||||
- 👍 **Powerful Video VAE**: **Wan-VAE** delivers exceptional efficiency and performance, encoding and decoding 1080P videos of any length while preserving temporal information, making it an ideal foundation for video and image generation.
|
||||
|
||||
## Video Demos
|
||||
|
||||
<div align="center">
|
||||
<video src="https://github.com/user-attachments/assets/4aca6063-60bf-4953-bfb7-e265053f49ef" width="70%" poster=""> </video>
|
||||
</div>
|
||||
|
||||
## 🔥 Latest News!!
|
||||
|
||||
* Feb 25, 2025: 👋 We've released the inference code and weights of Wan2.1.
|
||||
|
||||
|
||||
## 📑 Todo List
|
||||
- Wan2.1 Text-to-Video
|
||||
- [x] Multi-GPU Inference code of the 14B and 1.3B models
|
||||
- [x] Checkpoints of the 14B and 1.3B models
|
||||
- [x] Gradio demo
|
||||
- [ ] Diffusers integration
|
||||
- [ ] ComfyUI integration
|
||||
- Wan2.1 Image-to-Video
|
||||
- [x] Multi-GPU Inference code of the 14B model
|
||||
- [x] Checkpoints of the 14B model
|
||||
- [x] Gradio demo
|
||||
- [ ] Diffusers integration
|
||||
- [ ] ComfyUI integration
|
||||
|
||||
|
||||
## Quickstart
|
||||
|
||||
#### Installation
|
||||
Clone the repo:
|
||||
```
|
||||
git clone https://github.com/Wan-Video/Wan2.1.git
|
||||
cd Wan2.1
|
||||
```
|
||||
|
||||
Install dependencies:
|
||||
```
|
||||
# Ensure torch >= 2.4.0
|
||||
pip install -r requirements.txt
|
||||
```
|
||||
|
||||
|
||||
#### Model Download
|
||||
|
||||
| Models | Download Link | Notes |
|
||||
| --------------|-------------------------------------------------------------------------------|-------------------------------|
|
||||
| T2V-14B | 🤗 [Huggingface](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B) 🤖 [ModelScope](https://www.modelscope.cn/models/Wan-AI/Wan2.1-T2V-14B) | Supports both 480P and 720P
|
||||
| I2V-14B-720P | 🤗 [Huggingface](https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-720P) 🤖 [ModelScope](https://www.modelscope.cn/models/Wan-AI/Wan2.1-I2V-14B-720P) | Supports 720P
|
||||
| I2V-14B-480P | 🤗 [Huggingface](https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P) 🤖 [ModelScope](https://www.modelscope.cn/models/Wan-AI/Wan2.1-I2V-14B-480P) | Supports 480P
|
||||
| T2V-1.3B | 🤗 [Huggingface](https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B) 🤖 [ModelScope](https://www.modelscope.cn/models/Wan-AI/Wan2.1-T2V-1.3B) | Supports 480P
|
||||
|
||||
> 💡Note: The 1.3B model is capable of generating videos at 720P resolution. However, due to limited training at this resolution, the results are generally less stable compared to 480P. For optimal performance, we recommend using 480P resolution.
|
||||
|
||||
|
||||
Download models using huggingface-cli:
|
||||
```
|
||||
pip install "huggingface_hub[cli]"
|
||||
huggingface-cli download Wan-AI/Wan2.1-T2V-14B --local-dir ./Wan2.1-T2V-14B
|
||||
```
|
||||
|
||||
Download models using modelscope-cli:
|
||||
```
|
||||
pip install modelscope
|
||||
modelscope download Wan-AI/Wan2.1-T2V-14B --local_dir ./Wan2.1-T2V-14B
|
||||
```
|
||||
#### Run Text-to-Video Generation
|
||||
|
||||
This repository supports two Text-to-Video models (1.3B and 14B) and two resolutions (480P and 720P). The parameters and configurations for these models are as follows:
|
||||
|
||||
<table>
|
||||
<thead>
|
||||
<tr>
|
||||
<th rowspan="2">Task</th>
|
||||
<th colspan="2">Resolution</th>
|
||||
<th rowspan="2">Model</th>
|
||||
</tr>
|
||||
<tr>
|
||||
<th>480P</th>
|
||||
<th>720P</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
<tr>
|
||||
<td>t2v-14B</td>
|
||||
<td style="color: green;">✔️</td>
|
||||
<td style="color: green;">✔️</td>
|
||||
<td>Wan2.1-T2V-14B</td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td>t2v-1.3B</td>
|
||||
<td style="color: green;">✔️</td>
|
||||
<td style="color: red;">❌</td>
|
||||
<td>Wan2.1-T2V-1.3B</td>
|
||||
</tr>
|
||||
</tbody>
|
||||
</table>
|
||||
|
||||
|
||||
##### (1) Without Prompt Extention
|
||||
|
||||
To facilitate implementation, we will start with a basic version of the inference process that skips the [prompt extension](#2-using-prompt-extention) step.
|
||||
|
||||
- Single-GPU inference
|
||||
|
||||
```
|
||||
python generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
|
||||
```
|
||||
|
||||
If you encounter OOM (Out-of-Memory) issues, you can use the `--offload_model True` and `--t5_cpu` options to reduce GPU memory usage. For example, on an RTX 4090 GPU:
|
||||
|
||||
```
|
||||
python generate.py --task t2v-1.3B --size 832*480 --ckpt_dir ./Wan2.1-T2V-1.3B --offload_model True --t5_cpu --sample_shift 8 --sample_guide_scale 6 --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
|
||||
```
|
||||
|
||||
> 💡Note: If you are using the `T2V-1.3B` model, we recommend setting the parameter `--sample_guide_scale 6`. The `--sample_shift parameter` can be adjusted within the range of 8 to 12 based on the performance.
|
||||
|
||||
|
||||
- Multi-GPU inference using FSDP + xDiT USP
|
||||
|
||||
```
|
||||
pip install "xfuser>=0.4.1"
|
||||
torchrun --nproc_per_node=8 generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --dit_fsdp --t5_fsdp --ulysses_size 8 --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
|
||||
```
|
||||
|
||||
|
||||
##### (2) Using Prompt Extention
|
||||
|
||||
Extending the prompts can effectively enrich the details in the generated videos, further enhancing the video quality. Therefore, we recommend enabling prompt extension. We provide the following two methods for prompt extension:
|
||||
|
||||
- Use the Dashscope API for extension.
|
||||
- Apply for a `dashscope.api_key` in advance ([EN](https://www.alibabacloud.com/help/en/model-studio/getting-started/first-api-call-to-qwen) | [CN](https://help.aliyun.com/zh/model-studio/getting-started/first-api-call-to-qwen)).
|
||||
- Configure the environment variable `DASH_API_KEY` to specify the Dashscope API key. For users of Alibaba Cloud's international site, you also need to set the environment variable `DASH_API_URL` to 'https://dashscope-intl.aliyuncs.com/api/v1'. For more detailed instructions, please refer to the [dashscope document](https://www.alibabacloud.com/help/en/model-studio/developer-reference/use-qwen-by-calling-api?spm=a2c63.p38356.0.i1).
|
||||
- Use the `qwen-plus` model for text-to-video tasks and `qwen-vl-max` for image-to-video tasks.
|
||||
- You can modify the model used for extension with the parameter `--prompt_extend_model`. For example:
|
||||
```
|
||||
DASH_API_KEY=your_key python generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage" --use_prompt_extend --prompt_extend_method 'dashscope' --prompt_extend_target_lang 'ch'
|
||||
```
|
||||
|
||||
- Using a local model for extension.
|
||||
|
||||
- By default, the Qwen model on HuggingFace is used for this extension. Users can choose Qwen models or other models based on the available GPU memory size.
|
||||
- For text-to-video tasks, you can use models like `Qwen/Qwen2.5-14B-Instruct`, `Qwen/Qwen2.5-7B-Instruct` and `Qwen/Qwen2.5-3B-Instruct`.
|
||||
- For image-to-video tasks, you can use models like `Qwen/Qwen2.5-VL-7B-Instruct` and `Qwen/Qwen2.5-VL-3B-Instruct`.
|
||||
- Larger models generally provide better extension results but require more GPU memory.
|
||||
- You can modify the model used for extension with the parameter `--prompt_extend_model` , allowing you to specify either a local model path or a Hugging Face model. For example:
|
||||
|
||||
```
|
||||
python generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage" --use_prompt_extend --prompt_extend_method 'local_qwen' --prompt_extend_target_lang 'ch'
|
||||
```
|
||||
|
||||
##### (3) Runing local gradio
|
||||
|
||||
```
|
||||
cd gradio
|
||||
# if one uses dashscope’s API for prompt extension
|
||||
DASH_API_KEY=your_key python t2v_14B_singleGPU.py --prompt_extend_method 'dashscope' --ckpt_dir ./Wan2.1-T2V-14B
|
||||
|
||||
# if one uses a local model for prompt extension
|
||||
python t2v_14B_singleGPU.py --prompt_extend_method 'local_qwen' --ckpt_dir ./Wan2.1-T2V-14B
|
||||
```
|
||||
|
||||
|
||||
#### Run Image-to-Video Generation
|
||||
|
||||
Similar to Text-to-Video, Image-to-Video is also divided into processes with and without the prompt extension step. The specific parameters and their corresponding settings are as follows:
|
||||
<table>
|
||||
<thead>
|
||||
<tr>
|
||||
<th rowspan="2">Task</th>
|
||||
<th colspan="2">Resolution</th>
|
||||
<th rowspan="2">Model</th>
|
||||
</tr>
|
||||
<tr>
|
||||
<th>480P</th>
|
||||
<th>720P</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
<tr>
|
||||
<td>i2v-14B</td>
|
||||
<td style="color: green;">❌</td>
|
||||
<td style="color: green;">✔️</td>
|
||||
<td>Wan2.1-I2V-14B-720P</td>
|
||||
</tr>
|
||||
<tr>
|
||||
<td>i2v-14B</td>
|
||||
<td style="color: green;">✔️</td>
|
||||
<td style="color: red;">❌</td>
|
||||
<td>Wan2.1-T2V-14B-480P</td>
|
||||
</tr>
|
||||
</tbody>
|
||||
</table>
|
||||
|
||||
|
||||
##### (1) Without Prompt Extention
|
||||
|
||||
- Single-GPU inference
|
||||
```
|
||||
python generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image examples/i2v_input.JPG --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside."
|
||||
```
|
||||
|
||||
> 💡For the Image-to-Video task, the `size` parameter represents the area of the generated video, with the aspect ratio following that of the original input image.
|
||||
|
||||
|
||||
- Multi-GPU inference using FSDP + xDiT USP
|
||||
|
||||
```
|
||||
pip install "xfuser>=0.4.1"
|
||||
torchrun --nproc_per_node=8 generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image examples/i2v_input.JPG --dit_fsdp --t5_fsdp --ulysses_size 8 --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside."
|
||||
```
|
||||
|
||||
##### (2) Using Prompt Extention
|
||||
|
||||
|
||||
The process of prompt extension can be referenced [here](#2-using-prompt-extention).
|
||||
|
||||
Run with local prompt extention using `Qwen/Qwen2.5-VL-7B-Instruct`:
|
||||
```
|
||||
python generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image examples/i2v_input.JPG --use_prompt_extend --prompt_extend_model Qwen/Qwen2.5-VL-7B-Instruct --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside."
|
||||
```
|
||||
|
||||
Run with remote prompt extention using `dashscope`:
|
||||
```
|
||||
DASH_API_KEY=your_key python generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image examples/i2v_input.JPG --use_prompt_extend --prompt_extend_method 'dashscope' --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside."
|
||||
```
|
||||
|
||||
##### (3) Runing local gradio
|
||||
|
||||
```
|
||||
cd gradio
|
||||
# if one only uses 480P model in gradio
|
||||
DASH_API_KEY=your_key python i2v_14B_singleGPU.py --prompt_extend_method 'dashscope' --ckpt_dir_480p ./Wan2.1-I2V-14B-480P
|
||||
|
||||
# if one only uses 720P model in gradio
|
||||
DASH_API_KEY=your_key python i2v_14B_singleGPU.py --prompt_extend_method 'dashscope' --ckpt_dir_720p ./Wan2.1-I2V-14B-720P
|
||||
|
||||
# if one uses both 480P and 720P models in gradio
|
||||
DASH_API_KEY=your_key python i2v_14B_singleGPU.py --prompt_extend_method 'dashscope' --ckpt_dir_480p ./Wan2.1-I2V-14B-480P --ckpt_dir_720p ./Wan2.1-I2V-14B-720P
|
||||
```
|
||||
|
||||
|
||||
#### Run Text-to-Image Generation
|
||||
|
||||
Wan2.1 is a unified model for both image and video generation. Since it was trained on both types of data, it can also generate images. The command for generating images is similar to video generation, as follows:
|
||||
|
||||
##### (1) Without Prompt Extention
|
||||
|
||||
- Single-GPU inference
|
||||
```
|
||||
python generate.py --task t2i-14B --size 1024*1024 --ckpt_dir ./Wan2.1-T2V-14B --prompt '一个朴素端庄的美人'
|
||||
```
|
||||
|
||||
- Multi-GPU inference using FSDP + xDiT USP
|
||||
|
||||
```
|
||||
torchrun --nproc_per_node=8 generate.py --dit_fsdp --t5_fsdp --ulysses_size 8 --base_seed 0 --frame_num 1 --task t2i-14B --size 1024*1024 --prompt '一个朴素端庄的美人' --ckpt_dir ./Wan2.1-T2V-14B
|
||||
```
|
||||
|
||||
##### (2) With Prompt Extention
|
||||
|
||||
- Single-GPU inference
|
||||
```
|
||||
python generate.py --task t2i-14B --size 1024*1024 --ckpt_dir ./Wan2.1-T2V-14B --prompt '一个朴素端庄的美人' --use_prompt_extend
|
||||
```
|
||||
|
||||
- Multi-GPU inference using FSDP + xDiT USP
|
||||
```
|
||||
torchrun --nproc_per_node=8 generate.py --dit_fsdp --t5_fsdp --ulysses_size 8 --base_seed 0 --frame_num 1 --task t2i-14B --size 1024*1024 --ckpt_dir ./Wan2.1-T2V-14B --prompt '一个朴素端庄的美人' --use_prompt_extend
|
||||
```
|
||||
|
||||
|
||||
## Manual Evaluation
|
||||
|
||||
##### (1) Text-to-Video Evaluation
|
||||
|
||||
Through manual evaluation, the results generated after prompt extension are superior to those from both closed-source and open-source models.
|
||||
|
||||
<div align="center">
|
||||
<img src="assets/t2v_res.jpg" alt="" style="width: 80%;" />
|
||||
</div>
|
||||
|
||||
|
||||
##### (2) Image-to-Video Evaluation
|
||||
|
||||
We also conducted extensive manual evaluations to evaluate the performance of the Image-to-Video model, and the results are presented in the table below. The results clearly indicate that **Wan2.1** outperforms both closed-source and open-source models.
|
||||
|
||||
<div align="center">
|
||||
<img src="assets/i2v_res.png" alt="" style="width: 80%;" />
|
||||
</div>
|
||||
|
||||
|
||||
## Computational Efficiency on Different GPUs
|
||||
|
||||
We test the computational efficiency of different **Wan2.1** models on different GPUs in the following table. The results are presented in the format: **Total time (s) / peak GPU memory (GB)**.
|
||||
|
||||
|
||||
<div align="center">
|
||||
<img src="assets/comp_effic.png" alt="" style="width: 80%;" />
|
||||
</div>
|
||||
|
||||
> The parameter settings for the tests presented in this table are as follows:
|
||||
> (1) For the 1.3B model on 8 GPUs, set `--ring_size 8` and `--ulysses_size 1`;
|
||||
> (2) For the 14B model on 1 GPU, use `--offload_model True`;
|
||||
> (3) For the 1.3B model on a single 4090 GPU, set `--offload_model True --t5_cpu`;
|
||||
> (4) For all testings, no prompt extension was applied, meaning `--use_prompt_extend` was not enabled.
|
||||
|
||||
> 💡Note: T2V-14B is slower than I2V-14B because the former samples 50 steps while the latter uses 40 steps.
|
||||
|
||||
|
||||
## Community Contributions
|
||||
- [DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) provides more support for Wan, including video-to-video, FP8 quantization, VRAM optimization, LoRA training, and more. Please refer to [their examples](https://github.com/modelscope/DiffSynth-Studio/tree/main/examples/wanvideo).
|
||||
|
||||
-------
|
||||
|
||||
## Introduction of Wan2.1
|
||||
|
||||
**Wan2.1** is designed on the mainstream diffusion transformer paradigm, achieving significant advancements in generative capabilities through a series of innovations. These include our novel spatio-temporal variational autoencoder (VAE), scalable training strategies, large-scale data construction, and automated evaluation metrics. Collectively, these contributions enhance the model’s performance and versatility.
|
||||
|
||||
|
||||
##### (1) 3D Variational Autoencoders
|
||||
We propose a novel 3D causal VAE architecture, termed **Wan-VAE** specifically designed for video generation. By combining multiple strategies, we improve spatio-temporal compression, reduce memory usage, and ensure temporal causality. **Wan-VAE** demonstrates significant advantages in performance efficiency compared to other open-source VAEs. Furthermore, our **Wan-VAE** can encode and decode unlimited-length 1080P videos without losing historical temporal information, making it particularly well-suited for video generation tasks.
|
||||
|
||||
|
||||
<div align="center">
|
||||
<img src="assets/video_vae_res.jpg" alt="" style="width: 80%;" />
|
||||
</div>
|
||||
|
||||
|
||||
##### (2) Video Diffusion DiT
|
||||
|
||||
**Wan2.1** is designed using the Flow Matching framework within the paradigm of mainstream Diffusion Transformers. Our model's architecture uses the T5 Encoder to encode multilingual text input, with cross-attention in each transformer block embedding the text into the model structure. Additionally, we employ an MLP with a Linear layer and a SiLU layer to process the input time embeddings and predict six modulation parameters individually. This MLP is shared across all transformer blocks, with each block learning a distinct set of biases. Our experimental findings reveal a significant performance improvement with this approach at the same parameter scale.
|
||||
|
||||
<div align="center">
|
||||
<img src="assets/video_dit_arch.jpg" alt="" style="width: 80%;" />
|
||||
</div>
|
||||
|
||||
|
||||
| Model | Dimension | Input Dimension | Output Dimension | Feedforward Dimension | Frequency Dimension | Number of Heads | Number of Layers |
|
||||
|--------|-----------|-----------------|------------------|-----------------------|---------------------|-----------------|------------------|
|
||||
| 1.3B | 1536 | 16 | 16 | 8960 | 256 | 12 | 30 |
|
||||
| 14B | 5120 | 16 | 16 | 13824 | 256 | 40 | 40 |
|
||||
|
||||
|
||||
|
||||
##### Data
|
||||
|
||||
We curated and deduplicated a candidate dataset comprising a vast amount of image and video data. During the data curation process, we designed a four-step data cleaning process, focusing on fundamental dimensions, visual quality and motion quality. Through the robust data processing pipeline, we can easily obtain high-quality, diverse, and large-scale training sets of images and videos.
|
||||
|
||||
|
||||
|
||||
|
||||
##### Comparisons to SOTA
|
||||
We compared **Wan2.1** with leading open-source and closed-source models to evaluate the performace. Using our carefully designed set of 1,035 internal prompts, we tested across 14 major dimensions and 26 sub-dimensions. We then compute the total score by performing a weighted calculation on the scores of each dimension, utilizing weights derived from human preferences in the matching process. The detailed results are shown in the table below. These results demonstrate our model's superior performance compared to both open-source and closed-source models.
|
||||
|
||||
|
||||
|
||||
|
||||
## Citation
|
||||
If you find our work helpful, please cite us.
|
||||
|
||||
```
|
||||
@article{wan2.1,
|
||||
title = {Wan: Open and Advanced Large-Scale Video Generative Models},
|
||||
author = {Wan Team},
|
||||
journal = {},
|
||||
year = {2025}
|
||||
}
|
||||
```
|
||||
|
||||
## License Agreement
|
||||
The models in this repository are licensed under the Apache 2.0 License. We claim no rights over the your generate contents, granting you the freedom to use them while ensuring that your usage complies with the provisions of this license. You are fully accountable for your use of the models, which must not involve sharing any content that violates applicable laws, causes harm to individuals or groups, disseminates personal information intended for harm, spreads misinformation, or targets vulnerable populations. For a complete list of restrictions and details regarding your rights, please refer to the full text of the [license](LICENSE.txt).
|
||||
|
||||
|
||||
## Acknowledgements
|
||||
|
||||
We would like to thank the contributors to the [SD3](https://huggingface.co/stabilityai/stable-diffusion-3-medium), [Qwen](https://huggingface.co/Qwen), [umt5-xxl](https://huggingface.co/google/umt5-xxl), [diffusers](https://github.com/huggingface/diffusers) and [HuggingFace](https://huggingface.co) repositories, for their open research.
|
||||
|
||||
|
||||
|
||||
## Contact Us
|
||||
If you would like to leave a message to our research or product teams, feel free to join our [Discord](https://discord.gg/p5XbdQV7) or [WeChat groups](https://gw.alicdn.com/imgextra/i2/O1CN01tqjWFi1ByuyehkTSB_!!6000000000015-0-tps-611-1279.jpg)!
|
||||
227
generate.py
227
generate.py
@ -186,6 +186,11 @@ def _parse_args():
|
||||
type=float,
|
||||
default=5.0,
|
||||
help="Classifier free guidance scale.")
|
||||
parser.add_argument(
|
||||
"--device",
|
||||
type=str,
|
||||
default=None,
|
||||
help="Device to use for computation (mps, cpu).")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
@ -207,43 +212,33 @@ def _init_logging(rank):
|
||||
|
||||
|
||||
def generate(args):
|
||||
rank = int(os.getenv("RANK", 0))
|
||||
world_size = int(os.getenv("WORLD_SIZE", 1))
|
||||
local_rank = int(os.getenv("LOCAL_RANK", 0))
|
||||
device = local_rank
|
||||
_init_logging(rank)
|
||||
|
||||
# Handle both distributed and single-device scenarios
|
||||
if "RANK" in os.environ:
|
||||
# Distributed setup
|
||||
rank = int(os.getenv("RANK", 0))
|
||||
world_size = int(os.getenv("WORLD_SIZE", 1))
|
||||
local_rank = int(os.getenv("LOCAL_RANK", 0))
|
||||
device = torch.device(f"cuda:{local_rank}" if torch.cuda.is_available() else "cpu")
|
||||
init_logging(rank)
|
||||
else:
|
||||
# Single-device setup with MPS fallback
|
||||
rank = 0
|
||||
world_size = 1
|
||||
if args.device:
|
||||
device = torch.device(args.device)
|
||||
else:
|
||||
device = torch.device("cuda:0" if torch.cuda.is_available() else
|
||||
"mps" if torch.backends.mps.is_available() else
|
||||
"cpu")
|
||||
init_logging(rank)
|
||||
|
||||
# Ensure all torch operations use this device
|
||||
torch.set_default_device(device)
|
||||
|
||||
if args.offload_model is None:
|
||||
args.offload_model = False if world_size > 1 else True
|
||||
args.offload_model = True # Default to True for single device to save memory
|
||||
logging.info(
|
||||
f"offload_model is not specified, set to {args.offload_model}.")
|
||||
if world_size > 1:
|
||||
torch.cuda.set_device(local_rank)
|
||||
dist.init_process_group(
|
||||
backend="nccl",
|
||||
init_method="env://",
|
||||
rank=rank,
|
||||
world_size=world_size)
|
||||
else:
|
||||
assert not (
|
||||
args.t5_fsdp or args.dit_fsdp
|
||||
), f"t5_fsdp and dit_fsdp are not supported in non-distributed environments."
|
||||
assert not (
|
||||
args.ulysses_size > 1 or args.ring_size > 1
|
||||
), f"context parallel are not supported in non-distributed environments."
|
||||
|
||||
if args.ulysses_size > 1 or args.ring_size > 1:
|
||||
assert args.ulysses_size * args.ring_size == world_size, f"The number of ulysses_size and ring_size should be equal to the world size."
|
||||
from xfuser.core.distributed import (initialize_model_parallel,
|
||||
init_distributed_environment)
|
||||
init_distributed_environment(
|
||||
rank=dist.get_rank(), world_size=dist.get_world_size())
|
||||
|
||||
initialize_model_parallel(
|
||||
sequence_parallel_degree=dist.get_world_size(),
|
||||
ring_degree=args.ring_size,
|
||||
ulysses_degree=args.ulysses_size,
|
||||
)
|
||||
|
||||
if args.use_prompt_extend:
|
||||
if args.prompt_extend_method == "dashscope":
|
||||
@ -253,58 +248,44 @@ def generate(args):
|
||||
prompt_expander = QwenPromptExpander(
|
||||
model_name=args.prompt_extend_model,
|
||||
is_vl="i2v" in args.task,
|
||||
device=rank)
|
||||
device=device) # Use MPS/CPU device instead of rank
|
||||
else:
|
||||
raise NotImplementedError(
|
||||
f"Unsupport prompt_extend_method: {args.prompt_extend_method}")
|
||||
|
||||
cfg = WAN_CONFIGS[args.task]
|
||||
if args.ulysses_size > 1:
|
||||
assert cfg.num_heads % args.ulysses_size == 0, f"`num_heads` must be divisible by `ulysses_size`."
|
||||
|
||||
logging.info(f"Generation job args: {args}")
|
||||
logging.info(f"Generation model config: {cfg}")
|
||||
|
||||
if dist.is_initialized():
|
||||
base_seed = [args.base_seed] if rank == 0 else [None]
|
||||
dist.broadcast_object_list(base_seed, src=0)
|
||||
args.base_seed = base_seed[0]
|
||||
|
||||
if "t2v" in args.task or "t2i" in args.task:
|
||||
if args.prompt is None:
|
||||
args.prompt = EXAMPLE_PROMPT[args.task]["prompt"]
|
||||
logging.info(f"Input prompt: {args.prompt}")
|
||||
if args.use_prompt_extend:
|
||||
logging.info("Extending prompt ...")
|
||||
if rank == 0:
|
||||
prompt_output = prompt_expander(
|
||||
args.prompt,
|
||||
tar_lang=args.prompt_extend_target_lang,
|
||||
seed=args.base_seed)
|
||||
if prompt_output.status == False:
|
||||
logging.info(
|
||||
f"Extending prompt failed: {prompt_output.message}")
|
||||
logging.info("Falling back to original prompt.")
|
||||
input_prompt = args.prompt
|
||||
else:
|
||||
input_prompt = prompt_output.prompt
|
||||
input_prompt = [input_prompt]
|
||||
prompt_output = prompt_expander(
|
||||
args.prompt,
|
||||
tar_lang=args.prompt_extend_target_lang,
|
||||
seed=args.base_seed)
|
||||
if prompt_output.status == False:
|
||||
logging.info(
|
||||
f"Extending prompt failed: {prompt_output.message}")
|
||||
logging.info("Falling back to original prompt.")
|
||||
input_prompt = args.prompt
|
||||
else:
|
||||
input_prompt = [None]
|
||||
if dist.is_initialized():
|
||||
dist.broadcast_object_list(input_prompt, src=0)
|
||||
args.prompt = input_prompt[0]
|
||||
input_prompt = prompt_output.prompt
|
||||
args.prompt = input_prompt
|
||||
logging.info(f"Extended prompt: {args.prompt}")
|
||||
|
||||
logging.info("Creating WanT2V pipeline.")
|
||||
wan_t2v = wan.WanT2V(
|
||||
config=cfg,
|
||||
checkpoint_dir=args.ckpt_dir,
|
||||
device_id=device,
|
||||
rank=rank,
|
||||
t5_fsdp=args.t5_fsdp,
|
||||
dit_fsdp=args.dit_fsdp,
|
||||
use_usp=(args.ulysses_size > 1 or args.ring_size > 1),
|
||||
device_id=device, # Use MPS/CPU device instead of local_rank
|
||||
rank=0, # Single device, so use rank 0
|
||||
t5_fsdp=False, # Disable FSDP (not supported on MPS)
|
||||
dit_fsdp=False, # Disable FSDP (not supported on MPS)
|
||||
use_usp=False, # Disable Ulysses/ring parallelism (single device)
|
||||
t5_cpu=args.t5_cpu,
|
||||
)
|
||||
|
||||
@ -332,36 +313,30 @@ def generate(args):
|
||||
img = Image.open(args.image).convert("RGB")
|
||||
if args.use_prompt_extend:
|
||||
logging.info("Extending prompt ...")
|
||||
if rank == 0:
|
||||
prompt_output = prompt_expander(
|
||||
args.prompt,
|
||||
tar_lang=args.prompt_extend_target_lang,
|
||||
image=img,
|
||||
seed=args.base_seed)
|
||||
if prompt_output.status == False:
|
||||
logging.info(
|
||||
f"Extending prompt failed: {prompt_output.message}")
|
||||
logging.info("Falling back to original prompt.")
|
||||
input_prompt = args.prompt
|
||||
else:
|
||||
input_prompt = prompt_output.prompt
|
||||
input_prompt = [input_prompt]
|
||||
prompt_output = prompt_expander(
|
||||
args.prompt,
|
||||
tar_lang=args.prompt_extend_target_lang,
|
||||
image=img,
|
||||
seed=args.base_seed)
|
||||
if prompt_output.status == False:
|
||||
logging.info(
|
||||
f"Extending prompt failed: {prompt_output.message}")
|
||||
logging.info("Falling back to original prompt.")
|
||||
input_prompt = args.prompt
|
||||
else:
|
||||
input_prompt = [None]
|
||||
if dist.is_initialized():
|
||||
dist.broadcast_object_list(input_prompt, src=0)
|
||||
args.prompt = input_prompt[0]
|
||||
input_prompt = prompt_output.prompt
|
||||
args.prompt = input_prompt
|
||||
logging.info(f"Extended prompt: {args.prompt}")
|
||||
|
||||
logging.info("Creating WanI2V pipeline.")
|
||||
wan_i2v = wan.WanI2V(
|
||||
config=cfg,
|
||||
checkpoint_dir=args.ckpt_dir,
|
||||
device_id=device,
|
||||
rank=rank,
|
||||
t5_fsdp=args.t5_fsdp,
|
||||
dit_fsdp=args.dit_fsdp,
|
||||
use_usp=(args.ulysses_size > 1 or args.ring_size > 1),
|
||||
device_id=device, # Use MPS/CPU device instead of local_rank
|
||||
rank=0, # Single device, so use rank 0
|
||||
t5_fsdp=False, # Disable FSDP (not supported on MPS)
|
||||
dit_fsdp=False, # Disable FSDP (not supported on MPS)
|
||||
use_usp=False, # Disable Ulysses/ring parallelism (single device)
|
||||
t5_cpu=args.t5_cpu,
|
||||
)
|
||||
|
||||
@ -378,33 +353,61 @@ def generate(args):
|
||||
seed=args.base_seed,
|
||||
offload_model=args.offload_model)
|
||||
|
||||
if rank == 0:
|
||||
if args.save_file is None:
|
||||
formatted_time = datetime.now().strftime("%Y%m%d_%H%M%S")
|
||||
formatted_prompt = args.prompt.replace(" ", "_").replace("/",
|
||||
"_")[:50]
|
||||
suffix = '.png' if "t2i" in args.task else '.mp4'
|
||||
args.save_file = f"{args.task}_{args.size.replace('*','x') if sys.platform=='win32' else args.size}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}" + suffix
|
||||
|
||||
if "t2i" in args.task:
|
||||
logging.info(f"Saving generated image to {args.save_file}")
|
||||
cache_image(
|
||||
tensor=video.squeeze(1)[None],
|
||||
save_file=args.save_file,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
# Save output
|
||||
if args.save_file is None:
|
||||
formatted_time = datetime.now().strftime("%Y%m%d_%H%M%S")
|
||||
formatted_prompt = args.prompt.replace(" ", "_").replace("/", "_")[:50]
|
||||
suffix = '.png' if "t2i" in args.task else '.mp4'
|
||||
# Use Windows-compatible path if needed and include ulysses/ring parameters for distributed cases
|
||||
if sys.platform == 'win32':
|
||||
size_str = args.size.replace('*', 'x')
|
||||
else:
|
||||
logging.info(f"Saving generated video to {args.save_file}")
|
||||
cache_video(
|
||||
tensor=video[None],
|
||||
save_file=args.save_file,
|
||||
fps=cfg.sample_fps,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
size_str = args.size
|
||||
|
||||
# Single device (MPS/CPU) doesn't use ulysses/ring parallelism
|
||||
if args.device and (args.device == "mps" or args.device == "cpu"):
|
||||
args.save_file = f"{args.task}_{size_str}_{formatted_prompt}_{formatted_time}{suffix}"
|
||||
else:
|
||||
# Include parallelism parameters for distributed cases
|
||||
args.save_file = f"{args.task}_{size_str}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}{suffix}"
|
||||
|
||||
if "t2i" in args.task:
|
||||
logging.info(f"Saving generated image to {args.save_file}")
|
||||
cache_image(
|
||||
tensor=video.squeeze(1)[None],
|
||||
save_file=args.save_file,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
else:
|
||||
logging.info(f"Saving generated video to {args.save_file}")
|
||||
cache_video(
|
||||
tensor=video[None],
|
||||
save_file=args.save_file,
|
||||
fps=cfg.sample_fps,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
logging.info("Finished.")
|
||||
|
||||
if "t2i" in args.task:
|
||||
logging.info(f"Saving generated image to {args.save_file}")
|
||||
cache_image(
|
||||
tensor=video.squeeze(1)[None],
|
||||
save_file=args.save_file,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
else:
|
||||
logging.info(f"Saving generated video to {args.save_file}")
|
||||
cache_video(
|
||||
tensor=video[None],
|
||||
save_file=args.save_file,
|
||||
fps=cfg.sample_fps,
|
||||
nrow=1,
|
||||
normalize=True,
|
||||
value_range=(-1, 1))
|
||||
logging.info("Finished.")
|
||||
|
||||
if __name__ == "__main__":
|
||||
args = _parse_args()
|
||||
|
||||
@ -11,6 +11,6 @@ easydict
|
||||
ftfy
|
||||
dashscope
|
||||
imageio-ffmpeg
|
||||
flash_attn
|
||||
# flash_attn
|
||||
gradio>=5.0.0
|
||||
numpy>=1.23.5,<2
|
||||
|
||||
@ -7,11 +7,11 @@ wan_shared_cfg = EasyDict()
|
||||
|
||||
# t5
|
||||
wan_shared_cfg.t5_model = 'umt5_xxl'
|
||||
wan_shared_cfg.t5_dtype = torch.bfloat16
|
||||
wan_shared_cfg.t5_dtype = torch.float32
|
||||
wan_shared_cfg.text_len = 512
|
||||
|
||||
# transformer
|
||||
wan_shared_cfg.param_dtype = torch.bfloat16
|
||||
wan_shared_cfg.param_dtype = torch.float32
|
||||
|
||||
# inference
|
||||
wan_shared_cfg.num_train_timesteps = 1000
|
||||
|
||||
@ -14,7 +14,7 @@ i2v_14B.t5_tokenizer = 'google/umt5-xxl'
|
||||
|
||||
# clip
|
||||
i2v_14B.clip_model = 'clip_xlm_roberta_vit_h_14'
|
||||
i2v_14B.clip_dtype = torch.float16
|
||||
i2v_14B.clip_dtype = torch.float32
|
||||
i2v_14B.clip_checkpoint = 'models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth'
|
||||
i2v_14B.clip_tokenizer = 'xlm-roberta-large'
|
||||
|
||||
|
||||
@ -10,7 +10,7 @@ from torch.distributed.fsdp.wrap import lambda_auto_wrap_policy
|
||||
def shard_model(
|
||||
model,
|
||||
device_id,
|
||||
param_dtype=torch.bfloat16,
|
||||
param_dtype=torch.float32,
|
||||
reduce_dtype=torch.float32,
|
||||
buffer_dtype=torch.float32,
|
||||
process_group=None,
|
||||
|
||||
@ -151,9 +151,9 @@ def usp_attn_forward(self,
|
||||
seq_lens,
|
||||
grid_sizes,
|
||||
freqs,
|
||||
dtype=torch.bfloat16):
|
||||
dtype=torch.float32):
|
||||
b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim
|
||||
half_dtypes = (torch.float16, torch.bfloat16)
|
||||
half_dtypes = (torch.float16, torch.float32)
|
||||
|
||||
def half(x):
|
||||
return x if x.dtype in half_dtypes else x.to(dtype)
|
||||
|
||||
@ -63,7 +63,14 @@ class WanI2V:
|
||||
init_on_cpu (`bool`, *optional*, defaults to True):
|
||||
Enable initializing Transformer Model on CPU. Only works without FSDP or USP.
|
||||
"""
|
||||
self.device = torch.device(f"cuda:{device_id}")
|
||||
# Check if device_id is a torch.device instance
|
||||
if isinstance(device_id, torch.device):
|
||||
self.device = device_id
|
||||
elif device_id == "mps" or (isinstance(device_id, int) and device_id == -1):
|
||||
self.device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
|
||||
else:
|
||||
self.device = torch.device(f"cuda:{device_id}")
|
||||
|
||||
self.config = config
|
||||
self.rank = rank
|
||||
self.use_usp = use_usp
|
||||
@ -237,7 +244,7 @@ class WanI2V:
|
||||
y = self.vae.encode([
|
||||
torch.concat([
|
||||
torch.nn.functional.interpolate(
|
||||
img[None].cpu(), size=(h, w), mode='bicubic').transpose(
|
||||
img[None], size=(h, w), mode='bicubic').transpose(
|
||||
0, 1),
|
||||
torch.zeros(3, 80, h, w)
|
||||
],
|
||||
|
||||
@ -33,25 +33,26 @@ def flash_attention(
|
||||
causal=False,
|
||||
window_size=(-1, -1),
|
||||
deterministic=False,
|
||||
dtype=torch.bfloat16,
|
||||
dtype=torch.float32,
|
||||
version=None,
|
||||
):
|
||||
"""
|
||||
q: [B, Lq, Nq, C1].
|
||||
k: [B, Lk, Nk, C1].
|
||||
v: [B, Lk, Nk, C2]. Nq must be divisible by Nk.
|
||||
q_lens: [B].
|
||||
k_lens: [B].
|
||||
dropout_p: float. Dropout probability.
|
||||
softmax_scale: float. The scaling of QK^T before applying softmax.
|
||||
causal: bool. Whether to apply causal attention mask.
|
||||
window_size: (left right). If not (-1, -1), apply sliding window local attention.
|
||||
deterministic: bool. If True, slightly slower and uses more memory.
|
||||
dtype: torch.dtype. Apply when dtype of q/k/v is not float16/bfloat16.
|
||||
Flash attention implementation with fallback for CPU and MPS devices
|
||||
"""
|
||||
half_dtypes = (torch.float16, torch.bfloat16)
|
||||
half_dtypes = (torch.float16, torch.float32)
|
||||
assert dtype in half_dtypes
|
||||
assert q.device.type == 'cuda' and q.size(-1) <= 256
|
||||
assert q.size(-1) <= 256, "Sequence length exceeds the maximum limit."
|
||||
|
||||
# Add CPU/MPS fallback implementation
|
||||
if not (FLASH_ATTN_2_AVAILABLE or FLASH_ATTN_3_AVAILABLE) or q.device.type in ['cpu', 'mps']:
|
||||
# Implement standard attention for CPU/MPS
|
||||
return attention(q, k, v,
|
||||
q_lens=q_lens,
|
||||
k_lens=k_lens,
|
||||
dropout_p=dropout_p,
|
||||
softmax_scale=softmax_scale,
|
||||
causal=causal,
|
||||
window_size=window_size)
|
||||
|
||||
# params
|
||||
b, lq, lk, out_dtype = q.size(0), q.size(1), k.size(1), q.dtype
|
||||
@ -142,7 +143,7 @@ def attention(
|
||||
causal=False,
|
||||
window_size=(-1, -1),
|
||||
deterministic=False,
|
||||
dtype=torch.bfloat16,
|
||||
dtype=torch.float32,
|
||||
fa_version=None,
|
||||
):
|
||||
if FLASH_ATTN_2_AVAILABLE or FLASH_ATTN_3_AVAILABLE:
|
||||
|
||||
@ -16,7 +16,7 @@ def sinusoidal_embedding_1d(dim, position):
|
||||
# preprocess
|
||||
assert dim % 2 == 0
|
||||
half = dim // 2
|
||||
position = position.type(torch.float64)
|
||||
position = position.type(torch.float32)
|
||||
|
||||
# calculation
|
||||
sinusoid = torch.outer(
|
||||
@ -31,7 +31,7 @@ def rope_params(max_seq_len, dim, theta=10000):
|
||||
freqs = torch.outer(
|
||||
torch.arange(max_seq_len),
|
||||
1.0 / torch.pow(theta,
|
||||
torch.arange(0, dim, 2).to(torch.float64).div(dim)))
|
||||
torch.arange(0, dim, 2).to(torch.float32).div(dim)))
|
||||
freqs = torch.polar(torch.ones_like(freqs), freqs)
|
||||
return freqs
|
||||
|
||||
@ -49,7 +49,7 @@ def rope_apply(x, grid_sizes, freqs):
|
||||
seq_len = f * h * w
|
||||
|
||||
# precompute multipliers
|
||||
x_i = torch.view_as_complex(x[i, :seq_len].to(torch.float64).reshape(
|
||||
x_i = torch.view_as_complex(x[i, :seq_len].to(torch.float32).reshape(
|
||||
seq_len, n, -1, 2))
|
||||
freqs_i = torch.cat([
|
||||
freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1),
|
||||
|
||||
@ -61,7 +61,7 @@ class T5LayerNorm(nn.Module):
|
||||
def forward(self, x):
|
||||
x = x * torch.rsqrt(x.float().pow(2).mean(dim=-1, keepdim=True) +
|
||||
self.eps)
|
||||
if self.weight.dtype in [torch.float16, torch.bfloat16]:
|
||||
if self.weight.dtype in [torch.float16, torch.float32]:
|
||||
x = x.type_as(self.weight)
|
||||
return self.weight * x
|
||||
|
||||
@ -474,8 +474,8 @@ class T5EncoderModel:
|
||||
def __init__(
|
||||
self,
|
||||
text_len,
|
||||
dtype=torch.bfloat16,
|
||||
device=torch.cuda.current_device(),
|
||||
dtype=torch.float32,
|
||||
device='mps' if torch.backends.mps.is_available() else 'cpu',
|
||||
checkpoint_path=None,
|
||||
tokenizer_path=None,
|
||||
shard_fn=None,
|
||||
|
||||
@ -57,7 +57,7 @@ class WanT2V:
|
||||
t5_cpu (`bool`, *optional*, defaults to False):
|
||||
Whether to place T5 model on CPU. Only works without t5_fsdp.
|
||||
"""
|
||||
self.device = torch.device(f"cuda:{device_id}")
|
||||
self.device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
|
||||
self.config = config
|
||||
self.rank = rank
|
||||
self.t5_cpu = t5_cpu
|
||||
@ -85,16 +85,20 @@ class WanT2V:
|
||||
self.model.eval().requires_grad_(False)
|
||||
|
||||
if use_usp:
|
||||
from xfuser.core.distributed import \
|
||||
get_sequence_parallel_world_size
|
||||
|
||||
from .distributed.xdit_context_parallel import (usp_attn_forward,
|
||||
usp_dit_forward)
|
||||
for block in self.model.blocks:
|
||||
block.self_attn.forward = types.MethodType(
|
||||
usp_attn_forward, block.self_attn)
|
||||
self.model.forward = types.MethodType(usp_dit_forward, self.model)
|
||||
self.sp_size = get_sequence_parallel_world_size()
|
||||
try:
|
||||
from xfuser.core.distributed import get_sequence_parallel_world_size
|
||||
from .distributed.xdit_context_parallel import (usp_attn_forward,
|
||||
usp_dit_forward)
|
||||
for block in self.model.blocks:
|
||||
block.self_attn.forward = types.MethodType(
|
||||
usp_attn_forward, block.self_attn)
|
||||
self.model.forward = types.MethodType(usp_dit_forward, self.model)
|
||||
self.sp_size = get_sequence_parallel_world_size()
|
||||
except ImportError:
|
||||
# Fall back to non-USP mode on systems without xfuser
|
||||
logging.warning("xfuser package not found, disabling USP mode")
|
||||
self.sp_size = 1
|
||||
use_usp = False
|
||||
else:
|
||||
self.sp_size = 1
|
||||
|
||||
@ -260,7 +264,8 @@ class WanT2V:
|
||||
del sample_scheduler
|
||||
if offload_model:
|
||||
gc.collect()
|
||||
torch.cuda.synchronize()
|
||||
if torch.cuda.is_available():
|
||||
torch.cuda.synchronize()
|
||||
if dist.is_initialized():
|
||||
dist.barrier()
|
||||
|
||||
|
||||
@ -44,7 +44,7 @@ LM_EN_SYS_PROMPT = \
|
||||
'''1. For overly concise user inputs, reasonably infer and add details to make the video more complete and appealing without altering the original intent;\n''' \
|
||||
'''2. Enhance the main features in user descriptions (e.g., appearance, expression, quantity, race, posture, etc.), visual style, spatial relationships, and shot scales;\n''' \
|
||||
'''3. Output the entire prompt in English, retaining original text in quotes and titles, and preserving key input information;\n''' \
|
||||
'''4. Prompts should match the user’s intent and accurately reflect the specified style. If the user does not specify a style, choose the most appropriate style for the video;\n''' \
|
||||
'''4. Prompts should match the user's intent and accurately reflect the specified style. If the user does not specify a style, choose the most appropriate style for the video;\n''' \
|
||||
'''5. Emphasize motion information and different camera movements present in the input description;\n''' \
|
||||
'''6. Your output should have natural motion attributes. For the target category described, add natural actions of the target using simple and direct verbs;\n''' \
|
||||
'''7. The revised prompt should be around 80-100 words long.\n''' \
|
||||
@ -82,7 +82,7 @@ VL_EN_SYS_PROMPT = \
|
||||
'''1. For overly brief user inputs, reasonably infer and supplement details without changing the original meaning, making the image more complete and visually appealing;\n''' \
|
||||
'''2. Improve the characteristics of the main subject in the user's description (such as appearance, expression, quantity, ethnicity, posture, etc.), rendering style, spatial relationships, and camera angles;\n''' \
|
||||
'''3. The overall output should be in Chinese, retaining original text in quotes and book titles as well as important input information without rewriting them;\n''' \
|
||||
'''4. The prompt should match the user’s intent and provide a precise and detailed style description. If the user has not specified a style, you need to carefully analyze the style of the user's provided photo and use that as a reference for rewriting;\n''' \
|
||||
'''4. The prompt should match the user's intent and provide a precise and detailed style description. If the user has not specified a style, you need to carefully analyze the style of the user's provided photo and use that as a reference for rewriting;\n''' \
|
||||
'''5. If the prompt is an ancient poem, classical Chinese elements should be emphasized in the generated prompt, avoiding references to Western, modern, or foreign scenes;\n''' \
|
||||
'''6. You need to emphasize movement information in the input and different camera angles;\n''' \
|
||||
'''7. Your output should convey natural movement attributes, incorporating natural actions related to the described subject category, using simple and direct verbs as much as possible;\n''' \
|
||||
@ -93,7 +93,7 @@ VL_EN_SYS_PROMPT = \
|
||||
'''1. A Japanese fresh film-style photo of a young East Asian girl with double braids sitting by the boat. The girl wears a white square collar puff sleeve dress, decorated with pleats and buttons. She has fair skin, delicate features, and slightly melancholic eyes, staring directly at the camera. Her hair falls naturally, with bangs covering part of her forehead. She rests her hands on the boat, appearing natural and relaxed. The background features a blurred outdoor scene, with hints of blue sky, mountains, and some dry plants. The photo has a vintage film texture. A medium shot of a seated portrait.\n''' \
|
||||
'''2. An anime illustration in vibrant thick painting style of a white girl with cat ears holding a folder, showing a slightly dissatisfied expression. She has long dark purple hair and red eyes, wearing a dark gray skirt and a light gray top with a white waist tie and a name tag in bold Chinese characters that says "紫阳" (Ziyang). The background has a light yellow indoor tone, with faint outlines of some furniture visible. A pink halo hovers above her head, in a smooth Japanese cel-shading style. A close-up shot from a slightly elevated perspective.\n''' \
|
||||
'''3. CG game concept digital art featuring a huge crocodile with its mouth wide open, with trees and thorns growing on its back. The crocodile's skin is rough and grayish-white, resembling stone or wood texture. Its back is lush with trees, shrubs, and thorny protrusions. With its mouth agape, the crocodile reveals a pink tongue and sharp teeth. The background features a dusk sky with some distant trees, giving the overall scene a dark and cold atmosphere. A close-up from a low angle.\n''' \
|
||||
'''4. In the style of an American drama promotional poster, Walter White sits in a metal folding chair wearing a yellow protective suit, with the words "Breaking Bad" written in sans-serif English above him, surrounded by piles of dollar bills and blue plastic storage boxes. He wears glasses, staring forward, dressed in a yellow jumpsuit, with his hands resting on his knees, exuding a calm and confident demeanor. The background shows an abandoned, dim factory with light filtering through the windows. There’s a noticeable grainy texture. A medium shot with a straight-on close-up of the character.\n''' \
|
||||
'''4. In the style of an American drama promotional poster, Walter White sits in a metal folding chair wearing a yellow protective suit, with the words "Breaking Bad" written in sans-serif English above him, surrounded by piles of dollar bills and blue plastic storage boxes. He wears glasses, staring forward, dressed in a yellow jumpsuit, with his hands resting on his knees, exuding a calm and confident demeanor. The background shows an abandoned, dim factory with light filtering through the windows. There's a noticeable grainy texture. A medium shot with a straight-on close-up of the character.\n''' \
|
||||
'''Directly output the rewritten English text.'''
|
||||
|
||||
|
||||
@ -347,7 +347,7 @@ class QwenPromptExpander(PromptExpander):
|
||||
use_fast=True)
|
||||
self.model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
||||
self.model_name,
|
||||
torch_dtype=torch.bfloat16 if FLASH_VER == 2 else
|
||||
torch_dtype=torch.float32 if FLASH_VER == 2 else
|
||||
torch.float16 if "AWQ" in self.model_name else "auto",
|
||||
attn_implementation="flash_attention_2"
|
||||
if FLASH_VER == 2 else None,
|
||||
@ -363,6 +363,16 @@ class QwenPromptExpander(PromptExpander):
|
||||
device_map="cpu")
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
|
||||
|
||||
# Initialize device
|
||||
if isinstance(device, torch.device):
|
||||
self.device = device
|
||||
elif device == "mps" or (isinstance(device, str) and "mps" in device):
|
||||
self.device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
|
||||
elif isinstance(device, int) and device == -1:
|
||||
self.device = torch.device("cpu")
|
||||
else:
|
||||
self.device = torch.device(f"cuda:{device}" if isinstance(device, int) else device)
|
||||
|
||||
def extend(self, prompt, system_prompt, seed=-1, *args, **kwargs):
|
||||
self.model = self.model.to(self.device)
|
||||
messages = [{
|
||||
|
||||
@ -274,6 +274,12 @@ def get_video_reader_backend() -> str:
|
||||
def fetch_video(
|
||||
ele: dict,
|
||||
image_factor: int = IMAGE_FACTOR) -> torch.Tensor | list[Image.Image]:
|
||||
# Handle MPS device compatibility
|
||||
original_device = None
|
||||
if isinstance(ele.get("video"), torch.Tensor) and ele["video"].device.type == "mps":
|
||||
original_device = ele["video"].device
|
||||
ele["video"] = ele["video"].cpu()
|
||||
|
||||
if isinstance(ele["video"], str):
|
||||
video_reader_backend = get_video_reader_backend()
|
||||
video = VIDEO_READER_BACKENDS[video_reader_backend](ele)
|
||||
@ -324,6 +330,10 @@ def fetch_video(
|
||||
images.extend([images[-1]] * (nframes - len(images)))
|
||||
return images
|
||||
|
||||
# Return to original device if needed
|
||||
if original_device is not None and isinstance(video, torch.Tensor):
|
||||
video = video.to(original_device)
|
||||
|
||||
|
||||
def extract_vision_info(
|
||||
conversations: list[dict] | list[list[dict]]) -> list[dict]:
|
||||
|
||||
Loading…
Reference in New Issue
Block a user