ComfyUI-Qwen-Prompt-Rewrite

A ComfyUI custom node for Prompt Enhancer based on Qwen-Image-2.1. Supports both Text-to-Image (T2I) prompt generation and Image-to-Image (I2I) prompt rewriting with dynamic UI that supports 1-9 image connections.

中文指南 | English Guide

---

English Installation Guide

1. Install Custom Node

You can install this node via ComfyUI Manager by searching for this repository, or install it manually:

Navigate to your ComfyUI custom_nodes folder and clone this repository:

cd ComfyUI/custom_nodes
git clone https://github.com/kana112233/ComfyUI-kaola-qwen-image2.1-prompt-rewrite.git
cd ComfyUI-kaola-qwen-image2.1-prompt-rewrite
pip install -r requirements.txt

2. Download the Models

This node relies on standard HuggingFace directory formats, NOT single .safetensors files.

1. Go to HuggingFace or ModelScope to download the full repository:
   • For T2I (Text-only prompt enhancement): Qwen/Qwen-Image-2.1-PE-T2I
   • For I2I (Image reference prompt rewrite): Qwen/Qwen-Image-2.1-PE-I2I
   • Uncensored I2I (Abliterated, community version): base11231/Qwen-Image-2.1-PE-I2I-Abliterated
2. Create a folder named after the model (e.g. Qwen-Image-2.1-PE-I2I) in your ComfyUI/models/LLM directory.
3. Put all downloaded files (including config.json, tokenizer.json, etc.) into this newly created folder.

3. Usage & Examples

Restart ComfyUI and load the workflow files from the examples folder:

• t2i_prompt_enhancer_workflow.json: Pure text prompt expansion.
• i2i_edit_workflow.json: Image-based prompt editing (dynamically supports 1-9 images).
• i2t_caption_workflow.json: Reverse engineering image to prompt (extracts detailed prompt from an image).

Note: This model uses the newest qwen3_5 architecture. Ensure your transformers package is updated to the latest git branch as specified in requirements.txt (pip install git+https://github.com/huggingface/transformers.git).

---

中文安装指南

1. 安装自定义节点

推荐使用 ComfyUI Manager 搜索并直接安装。如果你想手动安装：

进入 ComfyUI 的 custom_nodes 目录，克隆此仓库并安装依赖：

cd ComfyUI/custom_nodes
git clone https://github.com/kana112233/ComfyUI-kaola-qwen-image2.1-prompt-rewrite.git
cd ComfyUI-kaola-qwen-image2.1-prompt-rewrite
pip install -r requirements.txt

2. 下载模型

本项目基于标准的开源大语言模型架构（调用 transformers），因此请勿仅下载单文件，必须下载完整的模型文件夹。

1. 前往 HuggingFace 或魔搭社区 (ModelScope) 下载完整的模型仓库里的所有文件：
   • 纯文本提示词扩写模型：Qwen/Qwen-Image-2.1-PE-T2I
   • 图像编辑重写模型：Qwen/Qwen-Image-2.1-PE-I2I
   • 无审查 I2I 模型（社区破限版）：base11231/Qwen-Image-2.1-PE-I2I-Abliterated
2. 在你的 ComfyUI/models/LLM 目录下，新建一个模型文件夹（例如 Qwen-Image-2.1-PE-I2I）。
3. 将下载的所有文件（包含 config.json, tokenizer.json 等全部文件）放进该文件夹中。

3. 工作流使用

重启 ComfyUI 后，可直接拖入本项目 examples 文件夹下的工作流体验：

• t2i_prompt_enhancer_workflow.json：纯文本生成高质量英文提示词。
• i2i_edit_workflow.json：看图重写提示词（前端自带动态连线功能，最高支持 1-9 张图片同时传入）。

注：由于该模型使用了极新的 qwen3_5 架构，必须确保你的 transformers 库是最新版（参考 requirements.txt，环境需执行 pip install git+https://github.com/huggingface/transformers.git）。否则会出现加载架构不认识的报错。