AgentSkillsCN

edit-with-comfyui

在ComfyUI中使用Qwen 2.5VL编辑图像，支持相机变化与图像分析。

SKILL.md
--- frontmatter
name: edit-with-comfyui
description: Edit images in ComfyUI using Qwen 2.5VL with support for camera changes and image analysis

Edit with ComfyUI Skill

Purpose

Specialized skill for editing existing images using ComfyUI and Qwen 2.5VL. Supports prompt-based edits (e.g., transforming objects, backgrounds) and camera changes (zoom, angle shifts).

Trigger

Use this skill when:

  • •User wants to modify an existing image (default: Flux Edit)
  • •User requests camera changes, stylization, or object manipulation
  • •User mentions "qwen" specifically (triggers Qwen Edit)

[!NOTE] Trigger words like "qwen edit:" or "qwen:" are used only for template selection and are automatically stripped from the prompt before being submitted to ComfyUI.

Workflow Templates

  1. •Flux Edit (Default): antigravity - movie - flux2edit.json
    • •Target Node for Prompt: 109 (CLIPTextEncode)
    • •Target Node for Image: 115 (LoadImageFromUrl)
  2. •Qwen Edit: movie 1.1 - qwen image edit antigravity.json
    • •Target Node for Prompt: 68 (TextEncodeQwenImageEditPlus)
    • •Target Node for Image: 117 or 41 (LoadImageFromUrl / LoadImage)

Prerequisites

  • •ComfyUI running at http://127.0.0.1:8188
  • •Workflow: movie 1.1 - qwen image edit antigravity.json
  • •Model: Qwen 2.5VL (installed in ComfyUI)

[!IMPORTANT] Performance: This workflow uses heavy models (Qwen 2.5VL) and can take several minutes (2-5 mins) to complete a single edit. The client has an extended 20-minute timeout to accommodate this.

How to Present Results

  1. •Images: Display edited results directly using ![Edited Image](/absolute/path/to/image.png).
  2. •Context: Explain the changes made (e.g., "Transformed turtle into a dragon, changed background to Rome").

Workflow Steps

Step 1: Analyze Image (Optional)

If the edit request is complex, first analyze the source image to identify key elements.

Step 2: Prepare Edit Prompt

  • •Subject Edits: Describe the transformation clearly (e.g., "transform the turtle into a majestic red dragon"). Narrative prompts work as well (e.g., "change the camera angle to medium closeup from behind").
  • •Background Edits: Specify new environments (e.g., "set in the ruins of ancient Rome").
  • •Camera Changes: Use identifying tags or narrative descriptions. Use the <sks> prefix for specific camera commands listed in node 90 if absolute precision is needed.

Step 3: Run Edit

Use the specialized client to run the edit workflow:

powershell
D:\ComfyUI7\python_embeded\python.exe ".agent\skills\edit-with-comfyui\scripts\comfyui_edit_client.py" --image "/path/to/source.png" --prompt "your edit prompt"

The client will automatically:

  1. •Copy the source image to the ComfyUI input directory.
  2. •Target node 41 for the image reference.
  3. •Target node 68 for the edit prompt.

Camera Changes Reference

Common triggers to include in prompts:

  • •<sks> front view
  • •<sks> low-angle shot
  • •<sks> close-up / <sks> wide shot
  • •<sks> zoom in / <sks> zoom out