comfyui-mcp/packs/qwen-image-edit-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-edit
display_name: Qwen-Image-Edit — Image Edit
family: qwen-image-edit
kind: img2img
description: >-
  Qwen-Image-Edit (Alibaba, Qwen2.5-VL-conditioned DiT) instruction-driven image
  editing — load an image, describe the change in natural language, get an edited
  result. Optionally stitch in a SECOND reference image (via ImageStitch) for
  compositional edits. Runs as a GGUF UNet + Qwen2.5-VL-7B text encoder with a
  4-step Lightning LoRA for fast turnaround. Single-purpose edit graph sliced
  from the qwen-image-edit monolith (one pipeline, no group toggles).
vram: "16GB+"
workflow: workflow.json
skill: qwen-image-edit
launch_args: []
sources:
  model: https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI
  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 "GGUF QWEN IMAGE EDIT".
  - Image-edit, not text-to-image — load a source image into the primary LoadImage
    and write the edit instruction in the TextEncodeQwenImageEdit prompt.
  - Optional SECOND IMAGE input is stitched in via ImageStitch for multi-image /
    compositional edits; leave it on the default image to edit a single picture.
  - Q8_0 UNet + Qwen2.5-VL-7B-Instruct-UD-Q8_0 are the shipped tiers; swap the
    UnetLoaderGGUF / CLIPLoaderGGUF widgets to Q4_K_S / Q5_K_S for lower VRAM.
  - BOTH text-encoder GGUFs (UD quant + mmproj-BF16 vision projector) must live in
    models/text_encoders/ or you hit a shape-mismatch error on load.
post_install:
  - Restart ComfyUI (or ComfyUI-Manager -> Install missing custom nodes for any
    node Python deps).
  - Load workflow.json.