Revise README for improved clarity and details

Updated README to enhance clarity on usage and features.
This commit is contained in:
overcloud81
2026-03-17 23:12:37 +08:00
committed by GitHub
parent b4d919e621
commit 3a15e8bb07
+3 -78
View File
@@ -1,17 +1,18 @@
# ComfyUI-ColorConsistencyHSL
**HSL Color Consistency Advanced**
A ComfyUI custom node for color/luminance consistency correction. Supports independent matching of Luminance (L), Hue (H), and Saturation (S) with multiple anchor modes and mask protection.
A ComfyUI custom node for color/luminance consistency correction. Works with any image editing, enhancement, or upscaling workflow – especially optimized to address color shift issues caused by models like Flux.2-Klein.
**Bilingual UI** – menu options automatically display in Chinese or English based on your ComfyUI language setting.
**HSL色彩一致性高级节点**
用于图像色彩/亮度一致性校正的 ComfyUI 自定义节点。支持亮度(L)、色相(H)、饱和度(S)的独立匹配,多种锚定模式和蒙版保护。
适用于任何图像编辑、增强或放大工作流的色彩/亮度一致性校正节点。特别针对 Flux.2-Klein 等模型导致的色彩偏移问题进行了优化。
**双语界面** – 根据 ComfyUI 语言设置,菜单选项自动显示中文或英文。
---
## Features / 功能亮点
- Universal applicability: works with any image editing/enhancement workflow / 通用性强:适用于任何图像编辑/增强工作流
- Independent control of L, H, S / 亮度、色相、饱和度独立控制
- Four anchor modes / 四种锚定模式:
- Statistical match / 统计匹配
@@ -35,79 +36,3 @@ Search for **"Color Consistency HSL"** in ComfyUI Manager and install directly.
```bash
cd ComfyUI/custom_nodes/
git clone https://github.com/yourusername/ComfyUI-ColorConsistencyHSL.git
```
### Option 3: Manual Download
Download the ZIP and extract into `ComfyUI/custom_nodes/` folder.
Restart ComfyUI. The node will appear in the `image/postprocessing` category as **“HSL Color Consistency Advanced”** (English) or **“色彩亮度一致性混合 (HSL高级版)”** (Chinese).
---
## Dependencies / 依赖
- `scipy` (optional, only for feathering; if missing, feathering will be disabled but node still works)
`scipy`(可选,仅用于羽化;如未安装,羽化功能禁用但节点仍可运行)
Install with:
```bash
pip install scipy
```
---
## Parameters / 参数说明
| Parameter | Description |
|-----------|-------------|
| `reference` | Reference image (original) / 参考图像(原图) |
| `target` | Target image (edited) / 目标图像(编辑后) |
| `mode` | Channels to match: luminance, hue, saturation, any combination / 要匹配的通道:亮度、色相、饱和度,任意组合 |
| `anchor_mode` | Matching mode: statistical, pixel-perfect, luminance stat+color exact, color exact / 匹配模式:统计匹配、像素级精确、亮度统计+色彩精确、色彩精确 |
| `luma_space` | Space for luminance matching: LAB L channel or linear RGB luminance / 亮度匹配空间:LAB L通道 或 线性RGB亮度 |
| `align_corners` | Align corners when scaling (reduces pixel shift) / 缩放时对齐四角(减少像素偏移) |
| `interpolation` | Interpolation method for scaling: bilinear / nearest / 缩放插值方式:双线性 / 最近邻 |
| `force_match_size` | Auto-scale reference to match target size if different / 若尺寸不同,自动缩放参考图以匹配目标 |
| `luma_strength` | Strength of luminance statistical matching (0=original, 1=full match) / 亮度统计匹配强度(0=保留原图,1=完全匹配) |
| `strength` | Global blend strength / 整图混合强度 |
| `protect_strength` | Proportion of original image retained in edit area (0=full match, 1=full original) / 编辑区域保留原图的比例(0=完全匹配,1=完全原图) |
| `feather_radius` | Feather radius for mask (pixels) / 蒙版羽化半径(像素) |
| `auto_mask` | Enable auto mask generation based on image difference / 启用基于图像差异的自动蒙版 |
| `mask_threshold` | Threshold for auto mask (0-1) / 自动蒙版阈值 |
| `external_mask` | Optional external mask (white=edit area, black=stable) / 可选外部蒙版(白色=编辑区域,黑色=稳定区域) |
---
## Usage Example / 使用示例
A typical workflow:
一个典型的工作流:
```
[Load Image] (original) → [Edit Model] → [Load Image] (edited)
↓
[original] → [Color Consistency HSL] (reference)
[edited] → [Color Consistency HSL] (target)
[optional mask] → [Color Consistency HSL] (external_mask)
↓
[Color Consistency HSL] → [Save Image]
```
Recommended settings for Klein edits with luminance preservation:
针对 Klein 增强且需保留亮度的推荐设置:
- `mode`: `luminance+hue+saturation`
- `anchor_mode`: `luminance stat + color exact`
- `luma_strength`: 0.5
- `protect_strength`: 0.3 (adjust based on edit area)
- `feather_radius`: 15
---
## License / 许可证
MIT License. See [LICENSE](LICENSE) for details.
## Author / 作者
**overcloud81** (@github @modelscope) with deepseek
Copyright © 2026 overcloud81. All rights reserved.