ComfyUI-PainterI2V

此节点由抖音博主：绘画小子 制作。

Wan2.2 图生视频增强节点，专门针对4步LoRA（如 lightx2v）的慢动作问题进行优化。

如果你用的是KJ的wan2.2-i2v工作流，请使用这个专门为KJ开发的节点，效果一致 (https://github.com/princepainter/ComfyUI-PainterI2VforKJ)

🛠 解决的问题

• ✅ 1.减少慢动作：提升运动幅度 15-50%

• ✅ 2.增强运镜：让运镜提示词更听话，运镜幅度更大

  和wan2.2官方节点效果对比

<table> <tr> <td><img src="image/wan2-2_00003.gif" alt="图1" width="300"></td> <td><img src="image/wan2-2_00004.gif" alt="图2" width="300"></td> <td><img src="image/wan2-2_00005.gif" alt="图3" width="300"></td> </tr> <tr> <td>motion_amplitude=1.2</td> <td>motion_amplitude=1.15</td> <td>motion_amplitude=1.15</td> </tr> <tr> <td><img src="image/wan2-2_00006.gif" alt="图4" width="300"></td> <td><img src="image/wan2-2_00007.gif" alt="图5" width="300"></td> <td><img src="image/wan2-2_00008.gif" alt="图6" width="300"></td> </tr> <tr> <td>motion_amplitude=1.5</td> <td>motion_amplitude=1.2</td> <td>motion_amplitude=1.15</td> </tr> </table>

以上视频对比效果，我是使用wan2.2官方模型+light2v 4步i2v lora V1.0做的测试（见下图），其他模型使用效果有待各位自行测试

节点特点

• 单帧输入优化：专为wan2.2单帧图生视频设计
• 即插即用：完全兼容原版 Wan2.2 工作流

📦 安装

方法 1: ComfyUI Manager（推荐）

1. 打开 ComfyUI Manager
2. 搜索 PainterI2V
3. 点击安装

方法 2: 手动安装

# 进入ComfyUI的custom_nodes目录
cd ComfyUI/custom_nodes

# 克隆仓库
git clone https://github.com/princepainter/ComfyUI-PainterI2V.git

重启ComfyUI

💡 使用方法

替换节点

在工作流中将 WanImageToVideo 替换为 PainterI2V

参数设置

• motion_amplitude: 1.15（推荐起始值）
• 其他参数与原版保持一致

场景参数推荐

| 运动类型 | 推荐参数 | 示例提示词 | |------------------|------------|------------------------| | 快速（跑步 / 跳跃） | 1.25–1.35 | "快速向前奔跑" | | 正常（走路 / 挥手） | 1.10–1.20 | "流畅地行走" | | 动态增强 | 1.00–1.10 | "略微增强动态和运镜" |

提示词优化

• 明确描述运动节奏，如 “快速奔跑”、“流畅行走”
• 避免模糊描述如 “移动”、“走动”

🔬 技术细节

| 参数值 | 运动提升 | 亮度变化 | 适用场景 | |--------|----------|----------|--------------| | 1.0（原版） | 0% | 无 | 和WAN原版节点无区别 | | 1.15（默认）| +15% | 无 | 通用场景 | | 1.3 | +30% | 无 | 体育运动 | | 1.5 | +50% | 无 | 极限运动 |

核心算法原理

• 亮度保护的运动缩放：放大运动向量前分离亮度均值
• 零 latent 初始化：严格保持 4 步 LoRA 的时序依赖链
• 参考帧增强：使用 reference_latents 保持主体一致性，不约束运动

⚡ 进阶技巧

• 最佳效果：配合强运动提示词使用
• 运动过快：每次减少 motion_amplitude 0.05
• 仍然偏慢：可适当增大到 1.4

🙏 致谢

• Wan2.2 团队：提供惊人的视频生成模型

• ComfyUI 社区：灵活的节点系统
  🙏如果这个项目对你有帮助，请给颗星 ⭐️ 支持一下！

• ComfyUI-Painterl2V

This node is created by Douyin creator: 绘画小子

PainterI2V node that specifically fixes the slow-motion issue in 4-step LoRAs (e.g., lightx2v).

The comparative effects in the above video are from my test using the official Wan2.2 model + Light2v 4-step I2V LoRA V1.0. The performance of other models awaits your own testing .

If you're using KJ's wan2.2-i2v workflow, use this node specifically developed for KJ — the effect is consistent.https://github.com/princepainter/ComfyUI-PainterI2VforKJ

🛠 Problems Solved

• ✅ Reduces Slow-Motion Drag: Increases motion amplitude by 15-50%
• ✅ Enhance camera movement: Make the camera movement prompts more responsive and increase the movement amplitude.
• ✅ Optimized for Single Frame: Designed specifically for single-frame image-to-video workflows
• ✅ Plug & Play: Fully compatible with original Wan2.2 workflows

📦 Installation

Method 1: ComfyUI Manager (Recommended)

1. Open ComfyUI Manager
2. Search for PainterI2V
3. Click Install

Method 2: Manual Installation

# Navigate to ComfyUI's custom_nodes directory
cd ComfyUI/custom_nodes

# Clone the repository
git clone https://github.com/princepainter/ComfyUI-PainterI2V.git

💡 Usage Guide

Replace the Node

In your workflow, replace WanImageToVideo with PainterI2V.

Parameter Settings

• motion_amplitude: 1.15 (recommended starting value)
• Keep all other parameters consistent with the original node.

Recommended Parameters by Scene

| Motion Type | Recommended Value | Example Prompt | |------------------------|-------------------|-----------------------------| | Fast (running / jumping) | 1.25–1.35 | "Running forward quickly" | | Normal (walking / waving)| 1.10–1.20 | "Walking smoothly" | | Motion Enhancement | 1.00–1.10 | "Slightly enhance motion and camera movement" |

Prompt Optimization Tips

• Clearly describe motion rhythm, e.g., “run fast”, “walk smoothly”
• Avoid vague terms like “move” or “walk around”

🔬 Technical Details

| Parameter | Motion Boost | Brightness Change | Best For | |-----------|--------------|-------------------|--------------------| | 1.0 (original) | 0% | None | No difference from the original WAN node | | 1.15 (default) | +15% | None | General use | | 1.3 | +30% | None | Sports action | | 1.5 | +50% | None | Extreme motion |

Core Algorithm Principles

• Brightness-Protected Motion Scaling: Separates luminance mean before scaling motion vectors
• Zero Latent Initialization: Strictly preserves the temporal dependency chain required by 4-step LoRAs
• Reference Frame Enhancement: Uses reference_latents to maintain subject consistency without constraining motion

⚡ Advanced Tips

• Best results: Use with strong motion-related prompts
• Motion too fast? Reduce motion_amplitude by 0.05 increments
• Still too slow? Try increasing up to 1.4

🙏 Acknowledgements

• Wan2.2 Team: For the amazing video generation model
• ComfyUI Community: For the flexible node system
• Contributors & testers: For helping refine this node

<div align="center"> If you find this project helpful, please give it a star ⭐ to show your support! </div> # Restart ComfyUI