comfyui-mcp/packs/z-image-turbo/pack.yaml

Local-first, agent-native control plane for ComfyUI — MCP server + autonomous sidebar agent that drives your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama (fully offline),

name: z-image-turbo
display_name: Z-Image Turbo
family: z-image
kind: txt2img
description: >-
  Z-Image Turbo — a 6B text-to-image model distilled for ~8-step turbo sampling,
  served as GGUF so it runs in under 8GB VRAM. This V2 ULTRA workflow adds a
  Fun-ControlNet-Union pass, a Detail Daemon detailer, SeedVarianceEnhancer, and
  txt2img / img2img / inpainting branches with rgthree group toggles.
vram: "<8GB"
workflow: workflow.json
skill: z-image-txt2img
launch_args: []
sources:
  mirror: https://huggingface.co/Aitrepreneur/FLX
  controlnet_aux: https://github.com/Fannovel16/comfyui_controlnet_aux
  detail_daemon: https://github.com/Jonseed/ComfyUI-Detail-Daemon
notes:
  - >-
    Run the generated installer from your ComfyUI root (the folder containing
    custom_nodes/ and models/).
  - >-
    The model is GGUF — it loads through ComfyUI-GGUF (UnetLoaderGGUF /
    CLIPLoaderGGUF). The workflow ships pinned to the Q8_0 UNet quant.
  - >-
    The installer menu offers a Q6_K quant, but no Q6_K GGUF exists on the
    Aitrepreneur/FLX mirror. Pick Q5_K_S (<8GB VRAM) or Q8_0 (best quality).
  - >-
    Pascal / Maxwell (GTX 10xx / 9xx) GPUs need the cu126 PyTorch build —
    reinstall torch from https://download.pytorch.org/whl/cu126 if generation
    fails on these cards.
  - >-
    "Torch not compiled with CUDA enabled" means a CPU-only torch is installed;
    reinstall the matching CUDA torch wheel (e.g. cu126/cu128) into the
    ComfyUI python_embeded / venv.
post_install:
  - >-
    Restart ComfyUI (or use ComfyUI-Manager → Install missing custom nodes for
    any node Python deps).
  - Load workflow.json.