comfyui-mcp/packs/z-image-base-txt2img/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-base-txt2img
display_name: Z-Image Base — Text to Image
family: z-image
kind: txt2img
description: >-
  Z-Image Base (Alibaba Tongyi, 6B S3-DiT base model — trainable, finetuning-friendly,
  more diverse styles than the distilled Turbo) text-to-image. Uses a Qwen text encoder
  (not CLIP/T5), GGUF-quantized to run in <8GB VRAM, paired with an 8-step distill LoRA.
  Single-purpose txt2img graph sliced from the z-image-base monolith (one pipeline, no
  group toggles), with optional film-grain / laplacian-sharpen post-processing and a
  Detailer Daemon detail pass.
vram: "<8GB"
workflow: workflow.json
skill: z-image-txt2img
launch_args: []
sources:
  model: https://huggingface.co/Aitrepreneur/FLX
  mirror: https://huggingface.co/Aitrepreneur/FLX
notes:
  - Run the generated installer from your ComfyUI root (folder with custom_nodes/
    and models/).
  - Sliced + un-bypassed from the monolith's "Z-IMAGE TEXT TO IMAGE".
  - Z-Image Base is the trainable, more-diverse counterpart to z-image-turbo.
  - The z-image VAE is saved as z-image-ae.safetensors (its weights differ from
    Flux's ae.safetensors despite the same size).
post_install:
  - Restart ComfyUI (or ComfyUI-Manager -> Install missing custom nodes for any
    node Python deps).
  - Load workflow.json.