Qwen-Image-Edit Skill
AI-powered image editing via tools/image_edit.py. The model depends on the deployment:
| Deployment |
Model |
License |
Defaults |
Max inputs |
--cloud modal (default) |
Qwen-Image-Edit-2511 |
Apache-2.0, commercial OK |
8 steps, guidance 1.0 |
3 images |
--cloud modal, app deployed with IMAGE_EDIT_MODEL=qwen-image-2.1 |
Qwen-Image-2.1 (2026-09, unified gen+edit, 7B DiT, RGBA) |
Qwen Research License -- non-commercial ("research or evaluation purposes only") |
40 steps, guidance 1.0 (no CFG) |
10 images |
--cloud runpod |
Qwen-Image-Edit-2511 + Lightning LoRA |
Apache-2.0, commercial OK |
8 steps, guidance 1.0 |
3 images |
The tool prints Model: and Steps: after each edit, so you can tell which model the Modal app
runs and how many steps it took. Without --steps, each model runs its own default above.
Opt in with IMAGE_EDIT_MODEL=qwen-image-2.1 uv run modal deploy docker/modal-image-edit/app.py;
deploy again without it to switch back.
Status: Evolving - learnings being captured as we experiment. The prompt
patterns and parameter findings in examples.md / parameters.md were measured
on 2511; re-check them on 2.1 before relying on the exact numbers.
Qwen-Image-2.1 specifics (Modal opt-in)
- Not step-distilled -- 40 steps is the model card default. The old 8-step
default was tuned for 2511's Lightning LoRA, which 2.1 doesn't have here, so expect
low step counts to cost quality (untested).
- Sampled without guidance.
--guidance 1.0 (default) is how the model is meant
to run. >1 switches on true CFG (blank negative prompt if you pass none) and
doubles the cost of every step.
- Multi-image: the first
--input is the image being edited ("image 1"); the
rest are references in order ("image 2", ...). Up to 10. Output size follows the
first image's aspect ratio at ~1MP.
- Local edits: draw a circle or paint an annotation on the image, or pass a
separate mask as an extra
--input, and say what to change there
(e.g. "Remove the watch inside the red circle").
- RGBA: transparent PNG inputs keep their alpha. For a cut-out, prompt
"This is an RGBA image with transparency. <subject>. The image has alpha channel and the background is transparent." -- the result comes back as an RGBA PNG only
when the model actually produced transparency.
When to Use This Skill
Use when the user wants to:
- Edit/transform photos while preserving identity
- Reframe cropped images (fix cut-off heads, etc.)
- Change clothing, add accessories
- Change pose (arm positions, hand placement)
- Apply style transfers (cyberpunk, anime, oil painting)
- Adjust lighting/color grading
- Add/remove objects
- Character transformations (Bond, Neo, etc.)
When NOT to Use
- Background replacement (single image) - creates cut-out artifacts, halos
- Face swapping - cannot preserve identity from reference
- Outpainting - can't extend canvas reliably
Use With Care
- Identity keep-clauses - naming features to keep ("keep his glasses and beard") ADDS them to people who lack
them. Use "Keep exactly the same person with an identical face and identical hair, do not add or remove glasses,
beard or stubble, photorealistic photograph" (costuming colleagues, 2026-11;
.claude/skills/parody-short/people.md).
For someone who does wear glasses, name them ("keeps wearing her round glasses"), or a new setting can drop them
- Multi-image compositing - CAN work with explicit identity anchors (see examples.md for prompt patterns). Requires describing distinctive features (hair texture/color, ethnicity, outfit) and using guidance ~2.0
- Camera angle changes - Inconsistent results. Vertical angles (low/high) work better than rotational (three-quarter view)
Quick Reference
# Basic edit
uv run tools/image_edit.py --input photo.jpg --prompt "Add sunglasses"
# With negative prompt (recommended)
uv run tools/image_edit.py --input photo.jpg \
--prompt "Reframe as portrait with full head visible" \
--negative "blur, distortion, artifacts"
# Style transfer
uv run tools/image_edit.py --input photo.jpg --style cyberpunk
# Background (use cautiously - often fails)
uv run tools/image_edit.py --input photo.jpg --background office
# Higher quality (RunPod / 2511 -- on Modal / 2.1 the 40-step default already is)
uv run tools/image_edit.py --input photo.jpg --prompt "..." --steps 16 --guidance 3.0 --cloud runpod
# Extract the subject onto a transparent background (Modal / 2.1)
uv run tools/image_edit.py --input product.jpg --prompt "This is an RGBA image with transparency. The product from the photo. The image has alpha channel and the background is transparent." --output product_cutout.png
# Multi-image composite (identity-preserving) -- settings tuned on RunPod / 2511
uv run tools/image_edit.py --input person.jpg background.jpg \
--prompt "The [ethnicity] [gender] with [hair description] from first image is now in [scene] from second image. Same [features], [outfit]." \
--negative "different ethnicity, different hair color, different face shape, generic stock photo" \
--steps 16 --guidance 2.0 --cloud runpod
# Multi-image composite on Modal / 2.1 -- up to 10 references, defaults usually suffice
uv run tools/image_edit.py --input person.jpg outfit.jpg shoes.jpg background.jpg \
--prompt "The person from image 1 wearing the outfit from image 2 and the shoes from image 3, standing in the scene from image 4. Same face and hair."
Key Files
prompting.md - Prompt patterns and structure
examples.md - Good/bad examples from experiments
parameters.md - Tuning steps, guidance, negative prompts
Tool Location
tools/image_edit.py - CLI wrapper for the Modal (docker/modal-image-edit/) and RunPod
(docker/runpod-qwen-edit/) endpoints
Related Docs
docs/qwen-edit-patterns.md - Character transformation patterns
.ai_dev/qwen-edit-research.md - Research notes
1---2name: qwen-edit3description: AI image editing prompting patterns for Qwen-Image (Qwen-Image-Edit-2511 by default, Qwen-Image-2.1 as a non-commercial Modal opt-in). Use when editing photos while preserving identity, reframing cropped images, changing clothing or accessories, adjusting poses, applying style transfers, or character transformations. Provides prompt patterns, parameter tuning, and examples.4metadata:5 source: https://skillmd.com/skills/majiayu000/qwen-edit6---78# Qwen-Image-Edit Skill910AI-powered image editing via `tools/image_edit.py`. The model depends on the deployment:1112| Deployment | Model | License | Defaults | Max inputs |13|-----------|-------|---------|----------|------------|14| `--cloud modal` (default) | Qwen-Image-Edit-2511 | Apache-2.0, commercial OK | 8 steps, guidance 1.0 | 3 images |15| `--cloud modal`, app deployed with `IMAGE_EDIT_MODEL=qwen-image-2.1` | **Qwen-Image-2.1** (2026-09, unified gen+edit, 7B DiT, RGBA) | Qwen Research License -- **non-commercial** ("research or evaluation purposes only") | 40 steps, guidance 1.0 (no CFG) | 10 images |16| `--cloud runpod` | Qwen-Image-Edit-2511 + Lightning LoRA | Apache-2.0, commercial OK | 8 steps, guidance 1.0 | 3 images |1718The tool prints `Model:` and `Steps:` after each edit, so you can tell which model the Modal app19runs and how many steps it took. Without `--steps`, each model runs its own default above.20Opt in with `IMAGE_EDIT_MODEL=qwen-image-2.1 uv run modal deploy docker/modal-image-edit/app.py`;21deploy again without it to switch back.2223**Status:** Evolving - learnings being captured as we experiment. The prompt24patterns and parameter findings in `examples.md` / `parameters.md` were measured25on 2511; re-check them on 2.1 before relying on the exact numbers.2627### Qwen-Image-2.1 specifics (Modal opt-in)2829- **Not step-distilled** -- 40 steps is the model card default. The old 8-step30 default was tuned for 2511's Lightning LoRA, which 2.1 doesn't have here, so expect31 low step counts to cost quality (untested).32- **Sampled without guidance.** `--guidance 1.0` (default) is how the model is meant33 to run. `>1` switches on true CFG (blank negative prompt if you pass none) and34 doubles the cost of every step.35- **Multi-image:** the first `--input` is the image being edited ("image 1"); the36 rest are references in order ("image 2", ...). Up to 10. Output size follows the37 first image's aspect ratio at ~1MP.38- **Local edits:** draw a circle or paint an annotation on the image, or pass a39 separate mask as an extra `--input`, and say what to change there40 (e.g. "Remove the watch inside the red circle").41- **RGBA:** transparent PNG inputs keep their alpha. For a cut-out, prompt42 `"This is an RGBA image with transparency. <subject>. The image has alpha channel43 and the background is transparent."` -- the result comes back as an RGBA PNG only44 when the model actually produced transparency.4546## When to Use This Skill4748Use when the user wants to:49- Edit/transform photos while preserving identity50- Reframe cropped images (fix cut-off heads, etc.)51- Change clothing, add accessories52- Change pose (arm positions, hand placement)53- Apply style transfers (cyberpunk, anime, oil painting)54- Adjust lighting/color grading55- Add/remove objects56- Character transformations (Bond, Neo, etc.)5758## When NOT to Use5960- **Background replacement (single image)** - creates cut-out artifacts, halos61- **Face swapping** - cannot preserve identity from reference62- **Outpainting** - can't extend canvas reliably6364## Use With Care6566- **Identity keep-clauses** - naming features to keep ("keep his glasses and beard") ADDS them to people who lack67 them. Use "Keep exactly the same person with an identical face and identical hair, do not add or remove glasses,68 beard or stubble, photorealistic photograph" (costuming colleagues, 2026-11; `.claude/skills/parody-short/people.md`).69 For someone who does wear glasses, name them ("keeps wearing her round glasses"), or a new setting can drop them70- **Multi-image compositing** - CAN work with explicit identity anchors (see examples.md for prompt patterns). Requires describing distinctive features (hair texture/color, ethnicity, outfit) and using guidance ~2.071- **Camera angle changes** - Inconsistent results. Vertical angles (low/high) work better than rotational (three-quarter view)7273## Quick Reference7475```bash76# Basic edit77uv run tools/image_edit.py --input photo.jpg --prompt "Add sunglasses"7879# With negative prompt (recommended)80uv run tools/image_edit.py --input photo.jpg \81 --prompt "Reframe as portrait with full head visible" \82 --negative "blur, distortion, artifacts"8384# Style transfer85uv run tools/image_edit.py --input photo.jpg --style cyberpunk8687# Background (use cautiously - often fails)88uv run tools/image_edit.py --input photo.jpg --background office8990# Higher quality (RunPod / 2511 -- on Modal / 2.1 the 40-step default already is)91uv run tools/image_edit.py --input photo.jpg --prompt "..." --steps 16 --guidance 3.0 --cloud runpod9293# Extract the subject onto a transparent background (Modal / 2.1)94uv run tools/image_edit.py --input product.jpg --prompt "This is an RGBA image with transparency. The product from the photo. The image has alpha channel and the background is transparent." --output product_cutout.png9596# Multi-image composite (identity-preserving) -- settings tuned on RunPod / 251197uv run tools/image_edit.py --input person.jpg background.jpg \98 --prompt "The [ethnicity] [gender] with [hair description] from first image is now in [scene] from second image. Same [features], [outfit]." \99 --negative "different ethnicity, different hair color, different face shape, generic stock photo" \100 --steps 16 --guidance 2.0 --cloud runpod101102# Multi-image composite on Modal / 2.1 -- up to 10 references, defaults usually suffice103uv run tools/image_edit.py --input person.jpg outfit.jpg shoes.jpg background.jpg \104 --prompt "The person from image 1 wearing the outfit from image 2 and the shoes from image 3, standing in the scene from image 4. Same face and hair."105```106107## Key Files108109- `prompting.md` - Prompt patterns and structure110- `examples.md` - Good/bad examples from experiments111- `parameters.md` - Tuning steps, guidance, negative prompts112113## Tool Location114115`tools/image_edit.py` - CLI wrapper for the Modal (`docker/modal-image-edit/`) and RunPod116(`docker/runpod-qwen-edit/`) endpoints117118## Related Docs119120- `docs/qwen-edit-patterns.md` - Character transformation patterns121- `.ai_dev/qwen-edit-research.md` - Research notes