ComfyUI Prompt Helper - Qwen3 Engineer

A simple ComfyUI custom node that loads the local GGUF version of Qwen3-4B-Z-Image-Engineer and expands short inputs into Z-Image Turbo–friendly positive prompts. Model card: https://huggingface.co/BennyDaBall/qwen3-4b-Z-Image-Engineer

Features

• Runs the GGUF text encoder locally via llama-cpp-python.
• Bundles the official system prompt focusing on positive constraints, texture detail, and camera settings.
• Outputs one enhanced prompt string to chain into your workflow.

Installation

1. Place this repo in ComfyUI/custom_nodes/ComfyUI-Prompt_Helper/.
2. Install dependencies: pip install -r requirements.txt.
3. Download the GGUF model file and put it under ComfyUI/models/text_encoders/ (or a subfolder).

Usage

1. Restart ComfyUI; the node category is QwenTextEngineer.
2. Pick gguf_name (auto-scans .gguf in models/text_encoders).
3. system_prompt is prefilled; adjust if needed. Enter your short description in prompt.
4. Run; it returns a rich positive prompt ready for downstream nodes.

Workflow

Parameters

• n_ctx: context length (default 4096).
• n_gpu_layers: -1 loads all layers to GPU when possible.
• max_new_tokens: generation length cap.
• temperature: sampling temperature (0–1).

FAQ

• If llama-cpp-python is missing, install dependencies then restart.
• If no GGUF file is found, ensure it exists under models/text_encoders.

Changelog

v1.1.0 (2024-12-17)

• Update precompiled files
• 新增 .gitignore 文件，忽略缓存目录
• 优化 安装体验，Windows 用户可直接使用本地 wheel 安装

v1.0.0

• 初始发布
• 支持本地 GGUF 模型加载
• 集成 Z-Image Turbo 专用系统提示词
• 自动扫描 models/text_encoders 目录

License

Apache-2.0 as per the upstream model.