Comfyui_DreamID-V_wrapper
A simple and efficient ComfyUI integration for DreamID-V, a model for identity-preserving video generation that animates reference images using motion from pose videos.
ComfyUI DreamID-V Wrapper
<a name="english"></a>
English
A simple and efficient ComfyUI integration for DreamID-V. DreamID-V is a model for Identity-Preserving Video Generation—it animates a reference image using motion from a pose video.
Features
- Simple Integration: Designed for easy use within ComfyUI.
- Selectable Precision: Choose between
bf16,fp16, etc. for best performance on your hardware. - Optimized: Built-in acceleration for faster generation.
- Flexible: Various options for face recognition precision and methods.
Installation
- Clone into
ComfyUI/custom_nodes. pip install -r requirements.txt.- Download Models:
dreamidv.pth->models/diffusion_models.Wan2.1_VAE.pth->models/vae.umt5-xxl-enc-bf16.pth->models/text_encoders.
Usage
- Load Model: Select precision (e.g.
bf16). - Extract Pose: Connect your video.
- Generate: Connect inputs and run.
Screenshots
![Workflow Example] <img width="4249" height="2225" alt="workflow (15)" src="https://github.com/user-attachments/assets/eaa3a81f-822f-4c75-bc5a-d3c83bb62e5e" />
<a name="中文说明"></a>
中文说明
DreamID-V 的 ComfyUI 简易整合包。 DreamID-V 是一款 ID保持视频生成 模型,它能根据一张参考图和一段骨架动作,生成保持人物身份且动作一致的视频。 主打简单易用,无需复杂配置即可体验高质量的人物视频生成。
主要功能
- 简单易用:专为 ComfyUI 设计的简洁节点,即插即用。
- 精度可选:支持
bf16、fp16等多种精度选择,适配不同显卡。 - 人脸识别:提供多种人脸识别精度和方式的选择。
- 极速生成:内置多种加速优化,生成速度快。
安装方法
- 在
custom_nodes目录下克隆本项目。 - 运行
pip install -r requirements.txt安装依赖。 - 模型准备:
dreamidv.pth放入models/diffusion_models。Wan2.1_VAE.pth放入models/vae。umt5-xxl-enc-bf16.pth放入models/text_encoders。
使用方法
- 加载模型:选择合适的精度(如 RTX 30/40系列推荐
bf16)。 - 提取姿态:使用姿态提取节点处理源视频。
- 生成视频:连接各节点并开始生成。
节点截图
![Workflow Screenshot] <img width="4249" height="2225" alt="workflow (15)" src="https://github.com/user-attachments/assets/eaa3a81f-822f-4c75-bc5a-d3c83bb62e5e" />
Credits
Wrapper for DreamID-V. Original code by Bytedance and The Alibaba Wan Team.