comfyui-mcp/packs/qwen-image-edit/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: qwen-image-edit
display_name: Qwen-Image-Edit
family: qwen
kind: edit
description: >-
  Qwen-Image-Edit — instruction-driven image editing (TextEncodeQwenImageEdit +
  ReferenceLatent) plus a manual-mask inpainting variant. Runs the Qwen-Image-Edit
  diffusion model as a city96 GGUF with the Qwen2.5-VL-7B-Instruct text encoder and a
  Lightning step-distill LoRA. Workflow ships the Q8_0 tier; Q4_K_S / Q5_K_S tiers are
  commented in manifest.yaml for lower-VRAM GPUs.
vram: "24GB+ (Q8_0, 20.3GB UNet — the tier this pack installs); Q5_K_S 12-16GB and Q4_K_S <12GB are commented out in manifest.yaml"
workflow: workflow.json
skill: qwen-image-edit
launch_args: []
sources:
  mirror: https://huggingface.co/Aitrepreneur/FLX
  model: https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI
  gguf: https://github.com/city96/ComfyUI-GGUF
notes:
  - >-
    Run the generated installer from your ComfyUI root (the folder containing
    custom_nodes/ and models/).
  - >-
    workflow-inpainting.json is the manual-mask "super inpainting" variant — paint a
    mask on the loaded image (SetLatentNoiseMask + GrowMaskWithBlur + ImageCompositeMasked)
    to restrict edits to the masked region. workflow.json is the standard instruction-edit graph.
  - >-
    "mat1 and mat2 shapes cannot be multiplied" fix: put BOTH Qwen2.5-VL text encoders
    (Qwen2.5-VL-7B-Instruct-UD-*.gguf AND Qwen2.5-VL-7B-Instruct-mmproj-BF16.gguf) in
    models/text_encoders/ per the city96 ComfyUI-GGUF guidance, then reload the workflow /
    restart ComfyUI so the loader re-scans the folder.
  - >-
    The Lightning 8-step LoRA needs an up-to-date ComfyUI — choose the Nightly ComfyUI
    build (ComfyUI-Manager > Update > switch to nightly) if step-distill sampling errors out.
  - >-
    "Torch not compiled with CUDA enabled" fix: reinstall the CUDA build of torch into the
    ComfyUI python (e.g. pip uninstall -y torch torchvision torchaudio, then reinstall from
    the matching https://download.pytorch.org/whl/cuXXX index).
post_install:
  - >-
    Restart ComfyUI (or use ComfyUI-Manager > Install missing custom nodes for any node
    Python deps).
  - >-
    Load workflow.json for instruction editing, or workflow-inpainting.json for manual-mask
    inpainting.