ComfyUI Z-Engineer

A custom node for ComfyUI that integrates a local LLM (via OpenAI-compatible API) to engineer optimal prompts for Z-Image Turbo workflows.

Features

• Local LLM Integration: Connects to your local LLM server (e.g., LM Studio, Oobabooga, vLLM) via standard OpenAI API format.
• System Prompting: Dedicated system prompt input to guide the LLM's behavior.
• Z-Image Optimization: Designed to work with models like Qwen3-4b-Z-Image-Engineer.

Installation

1. Clone this repository into your ComfyUI custom_nodes directory:

   cd ComfyUI/custom_nodes
   git clone https://github.com/BennyDaBall/ComfyUI-Z-Engineer.git
   

2. Restart ComfyUI.

Usage

1. Find the node under Z-Engineer > Z-Engineer.
2. Connect the output prompt to your CLIP Text Encode node.
3. Configuration:
   • Input Prompt: Your raw idea or base prompt.
   • System Prompt: Instructions for the LLM (System prompt supplied in HF repo!).
   • API URL: Your local LLM endpoint (default: http://localhost:1234/v1).
   • Model: The model name string (default: local-model).
   • Seed/Temperature: Standard LLM generation parameters.

Recommended Model

For best results, use the Z-Image Engineer model: BennyDaBall/qwen3-4b-Z-Image-Engineer

Requirements

• A running local LLM server compatible with OpenAI's chat/completions endpoint.
• ComfyUI installed.