Compatible with various models including Animagine, Boogu, Erine, Flux2, K9B, Ideogram, Lumina, Pony, Qwen, ZIT, and KREA2.

Local LLM prompt Enhancer support Qwen 3.8 9B (Extreme Fast)

For Krea 2 image reverse analysis, only take 12sec

The only downside is that the node descriptions are in Chinese, but a translation reference has already been pasted into the workflow.

唯一缺點是node的說明是中文, 在Workflow中已經貼上對照的翻譯

Beside the main model, make sure you had download the mmproj.

Low VRAM user may choose Q4_K_M

Put the main model & mmproj files (recommended format .gguf) into ComfyUI/models/LLM/

除了主模型之外，請確保您已下載 mmproj。

低顯存（VRAM）用戶可以選擇 Q4_K_M。

請將主模型與 mmproj 檔案（建議格式為 .gguf）放入 ComfyUI/models/LLM/ 目錄中。

### Image Reverse Engineering (Interrogator):

1. Select the model inside the purple node. (non chinese user please choose EN)

2. Connect the image.

3. Click RUN.

### Prompt Enhancer:

1. Enter a simple prompt in the green node.

2. No need to connect an image.

3. Click RUN.

### 反推圖片:

1. 在紫色Node內選取模型

2. 連接圖片

3. 按RUN

### Prompt Enhancer

1. 在綠色Node輸入簡單Prompt

2. 不用連接圖片

3. 按RUN

All related link had been written in the workflow.

If you meet trouble in the installation of llama cpp python, read this:👇

https://civarchive.com/articles/33986/how-to-install-llama-cpp-python-in-comfyui

All Credit belongs to node creator.

Special Thanks:👏