Extensions/Comfyui_DreamID-V_wrapper
ComfyUI Extension

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.

By TTPlanetPig·Created 7 months ago·Updated 7 months ago· 39
TTPlanetPig/Comfyui_DreamID-V_wrapper
Nodes
On cloudLocal install
Stars39
Updated7 months ago
Readme

ComfyUI DreamID-V Wrapper

中文版 | English


<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

  1. Clone into ComfyUI/custom_nodes.
  2. pip install -r requirements.txt.
  3. Download Models:
    • dreamidv.pth -> models/diffusion_models.
    • Wan2.1_VAE.pth -> models/vae.
    • umt5-xxl-enc-bf16.pth -> models/text_encoders.

Usage

  1. Load Model: Select precision (e.g. bf16).
  2. Extract Pose: Connect your video.
  3. 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 bf16fp16 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

  1. custom_nodes 目录下克隆本项目。
  2. 运行 pip install -r requirements.txt 安装依赖。
  3. 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

  1. 加载模型:选择合适的精度(如 RTX 30/40系列推荐 bf16)。
  2. 提取姿态:使用姿态提取节点处理源视频。
  3. 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.