ComfyUI-MagicClip_Strength

★ 2

文本编码器SDXL强度调节layer_idx 支持

在 ComfyUI 中分别调整 SDXL 的 clip_g 与 clip_l 文本编码器强度，支持正向 layer_idx 与正负混合，提升文本到图像对齐。

💡 在 ComfyUI 中微调 clip_g/clip_l 强度以改善文本到图像一致性。

🍴 1 Forks💻 Python🔄 2024-09-22

🔗 GitHub 原文

📦

网盘下载

复制链接后前往夸克网盘下载

https://pan.quark.cn/s/e98a62d17551

📄 README

🎨🖼️ ComfyUI MagicClip_Strength for SDXL 🖼️🎨

🎨 Regular Clip Text Encode (ZZzzz)

✨ Introduction

This project allows you to adjust SDXL’s two text encoder’s strengths individually for clip_g (ViT-bigG) and clip_l (CLIP-ViT-L) within ComfyUI. (And other adjustments)

🔧 Features

🎯 Clip Text Encoding: Adjust clip_g (global) and clip_l (local) strengths for better text-to-image alignment.

🖼️ Enhanced Layer_idx values: Specify positive layer_idx values. (ComfyUI usually just only supports negative values.) You can mix and match with + and – values on positive and negative prompt.

🛠️ Installation Instructions

Download ZIP and Extract:

Download the ZIP and extract it to:

“`

/ComfyUI/custom_nodes/ComfyUI-MagicClip_Strength

“`

Or Clone the Repository:

Clone the repository directly into your ComfyUI custom nodes directory:

“`sh

git clone https://github.com/311-code/ComfyUI-MagicClip_Strength

“`

Find the Node:

In ComfyUI, type SDXL to find the node labeled:

“`

311_code SDXL Clip Text Encode

“`

No requirements.txt file is needed.

💡 How to Use

After installation, use the node to adjust Clip strength directly in your workflows.

For more refined control over SDXL models, experiment with clip_g and clip_l strengths and positive and negative values, layer_idx, and size_cond_factor. (as shown in example image)

PS. I also highly trying out a ‘saveclip’ node on your favorite SDXL checkpoint, then use the newly saved custom clip_l in dualclip loader to mix and match clip_l with various models. (use this node and adjust strength)

Please note that using all positive layer_idx values for both the positive and negative prompts can be hit or miss. For example, here is the default ComfyUI purple galaxy glass bottle prompt with values +8 for pos/neg prompts. I recommend either mixing positive and negative values or using only negative values for better prompt adherence.

💖 Support Me

Creating and maintaining this project takes time and effort. If you find it helpful, consider supporting me:

🤝 Contributing

Contributions are welcome! If you have any ideas, suggestions, or bug reports, please feel free to open an issue or submit a pull request.

📜 License

This project is licensed under the Apache License – see the LICENSE file for details.