DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to edit and generate Qwen-Image pictures from a reference with Craft Skills

Use qwen-image-gen to rewrite prompts, edit regions, keep one character across a series, and check the original file.

Short answer: git clone https://github.com/ZSeven-W/craft-skills.git; then cp -R craft-skills/skills/qwen-image-gen "${CODEX_HOME:-$HOME/.codex}/skills/"; Restart or reload your agent session, and confirm your existing Qwen connection is available to the agent. All steps.

  • Content
  • Advanced
  • 1 hour
  • From the README
  • Codex

Credentials. The skill runs on your existing Qwen connection, so keep access keys in your own setup and out of prompts and shared files.

Who it is forA creator who already runs a Qwen-Image setup and needs portraits, edits, or a same-character series from a reference photo.

The problemQwen-Image results often drift off the brief, and sizes and reference relationships are easy to get wrong.

The outcomeA set of generated or edited images with the prompt rewrite, reference relationship, and size settings recorded, each checked as an original file.

What you need first

  • git, to clone the repository
  • An agent that reads skills from the Codex skills directory (CODEX_HOME or ~/.codex/skills)
  • Image generation available to the agent for the image skills
  • An existing Qwen connection for qwen-image-gen
  • Python 3.10 or later, only if you run the release check

Steps

  1. git clone https://github.com/ZSeven-W/craft-skills.git
  2. cp -R craft-skills/skills/qwen-image-gen "${CODEX_HOME:-$HOME/.codex}/skills/" (inferred, check the README)
  3. Restart or reload your agent session, and confirm your existing Qwen connection is available to the agent.
  4. Ask the agent for the series, and review each output file at full resolution.

Prompt for your agent

Use the qwen-image-gen skill from the craft-skills repo to create three separate portraits of the person in <reference image>, one action each, using my existing Qwen connection. Return the three original image files and the rewritten prompt for each. Verify that each file keeps the same identity and the requested frame size.

Caveat. The README says the sample images and offline tests do not guarantee every generation meets the brief, and that a complex chair pose still failed in its comparison.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for ZSeven-W/craft-skills and "Edit and generate Qwen-Image pictures from a reference"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer