Du hast einen ComfyUI-Workflow? Erstelle direkt im Browser eine verständliche Workflow-Dokumentation.Workflow verständlich machen.Jetzt ausprobieren ↗
Custom node for ComfyUI to engineer prompts for Z-Image Turbo
Paketprofil
README
Run Z-Image-Engineer-V6 fully inside ComfyUI: load the Qwen3-4B prompt model from sharded HuggingFace safetensors , use it as the , and use the as a with a live preview on the node. No LM Studio or external server required.
Quellen
3 QuellenVerifiziert 13. Sept.
Quellenauszüge
1 AuszugSource context: Repo BennyDaBall930/ComfyUI-Z-Engineer
Also runs the fast 1.2B LFM2.5-Z-Image-Engineer as a standalone prompt enhancer (GGUF or safetensors, on CUDA / ROCm / MPS / CPU via plain torch — no llama.cpp build needed). See LFM2.5 Engineer below.
Z-Engineer GGUF loader feeding the local prompt enhancer, with keep_terms set and the enhanced prompt previewed on the node
| Node | What it does |
|---|---|
| Z-Engineer CLIP Loader (Safetensors / Shards) | Loads a single .safetensors file or a sharded HF folder (model-00001-of-00003.safetensors + index) as a Z-Image text encoder. This is how you load the 3-piece Z-Image-Engineer-V6 release directly. |
| Z-Engineer CLIP Loader (GGUF) | Loads a llama.cpp-style Qwen3 GGUF (Q3_K_M ... F16) as a Z-Image text encoder. Uses ComfyUI-GGUF for on-the-fly dequant when installed; otherwise falls back to FP16 dequant at load time. |
| Z-Engineer Prompt Enhancer (Local) | Takes the loaded CLIP and generates the polished V6 prompt in-process (ComfyUI's own model management — no server). The enhanced prompt is previewed on the node and returned as a STRING. |
| Z-Engineer LFM2.5 Enhancer Loader (GGUF / Safetensors) | Loads LFM2.5-1.2B-Z-Image-Engineer (any GGUF quant, or the HF safetensors/ folder) as a standalone prompt-writer LLM. Runs on plain torch/transformers — CUDA, ROCm, MPS or CPU. Not a CLIP. |
| Z-Engineer Prompt Enhancer (LFM2.5 Local) | Same enhancer controls/preview as the Local node, but driven by the LFM2.5 model (~3x faster than Qwen3-4B). Wire its STRING output into CLIP Text Encode. |
| Z-Engineer Prompt Enhancer (API) |
Search for ComfyUI Z-Engineer in ComfyUI Manager, or:
comfy node install comfyui-z-engineer
cd ComfyUI/custom_nodes
git clone https://github.com/BennyDaBall930/ComfyUI-Z-Engineer.git
pip install -r ComfyUI-Z-Engineer/requirements.txt
Restart ComfyUI.
Put the model under ComfyUI/models/text_encoders/:
Sharded safetensors (the published V6 release):
ComfyUI/models/text_encoders/Z-Image-Engineer-V6/
├── model-00001-of-00003.safetensors
├── model-00002-of-00003.safetensors
├── model-00003-of-00003.safetensors
└── model.safetensors.index.json
Download from BennyDaBall/Z-Image-Engineer-V6. The folder shows up in the loader dropdown as Z-Image-Engineer-V6/.
GGUF (recommended for low VRAM):
ComfyUI/models/text_encoders/Z-Image-Engineer-V6-Q4_K_M.gguf
Download any quant from BennyDaBall/Z-Image-Engineer-V6-GGUF (Q4_K_M is a good default; F16 for maximum fidelity). Installing ComfyUI-GGUF is strongly recommended — the quant then stays quantized in VRAM.
clip into your normal CLIP Text Encode for theKnoten in diesem Paket
6 Knoten (Plural)Verifiziert 13. Sept.
Legacy path: OpenAI-compatible /chat/completions (LM Studio, llama.cpp server, Ollama). Kept for users who prefer an external server. |