fal-qwen-image-edit

Required inputs

• A text prompt describing the desired result.
• One or more reference images (--ref) as local files or URLs.
  • Typically: 1 “base” image to edit + optional extra reference/style images.
• A fal.ai API key in workers.jsonc (fal.key).

Optional inputs

• Output sizes / aspect ratios:
  • --image-size <preset> (repeatable) or --sizes a,b,c (comma-separated presets)
  • --size <WxH> (repeatable pixel-size hint; mapped to the closest supported preset)
  • Presets: square_hd, square, portrait_4_3, landscape_4_3, portrait_16_9, landscape_16_9
• --negative-prompt, --seed, --num-images, and other quality controls supported by the script.

Workflow

1. Prepare inputs

• Collect the base image and any extra references (PNG/JPEG recommended).
• Write a prompt that is specific and testable:
  • Subject: what should be preserved/changed from the base image.
  • Composition: angle, crop, framing, background.
  • Lighting + color: time of day, soft/hard light, mood.
  • Style: photoreal, studio product shot, cinematic, etc.
• If you need to avoid certain elements (logos, text, hands, extra objects), use --negative-prompt.

2. Run the generator

• Use the bundled script to call fal.ai.
• Final deliverables go to artifacts/; request/response logs and intermediate files stay in logs/.
  • Basic:
    • node skills/car-ads-designer/fal-qwen-image-edit/scripts/generate.mjs --prompt "..." --ref path/to/base.jpg
  • Multiple sizes (one call, multiple requests):
    • node skills/car-ads-designer/fal-qwen-image-edit/scripts/generate.mjs --prompt "..." --ref base.jpg --image-size square_hd --image-size landscape_16_9
  • Pixel-size hints (mapped to presets):
    • node skills/car-ads-designer/fal-qwen-image-edit/scripts/generate.mjs --prompt "..." --ref base.jpg --size 1024x1024 --size 1920x1080

3. Stop and request review (required)

• After generation completes, do not take any further actions (no auto-retouching, no upscaling, no additional variants) until a human or designated reviewer confirms the outputs are acceptable.

4. Review outputs

• Inspect the final images under artifacts/... first.
• If something looks off, use the working folder in logs/... to debug (it includes request/response payloads).

Outputs

• Working folder under logs/fal-qwen-image-edit/<run-id>/ containing inputs and full request/response logs.
• Final folder under artifacts/fal-qwen-image-edit/<run-id>/ containing:
  • README.md (what was generated + where the working logs are)
  • One or more final images (and per-size subfolders when multiple sizes are requested)

Definition of done

• The script completes without errors and writes at least one image file under artifacts/.
• The output matches the prompt intent and aligns with the provided reference image(s).
• A human (or designated agent reviewer) has reviewed the outputs and explicitly approved any next steps.

Safety / quality checklist

• Do not commit API keys; keep them in local-only workers.jsonc and avoid printing them.
• Avoid generating copyrighted logos/marks you don’t have rights to use.
• Treat rendered text as unreliable; avoid high-stakes text (prices, legal claims) unless you will manually verify/correct.
• Don’t include sensitive personal data in prompts or images unless you have permission and a clear need.