Edit and retouch · Alibaba (Qwen)
Qwen Image (edit)
Instruction-following edits that keep the rest of the image intact. Qwen Image 2 is the Studio's first "Automatic" choice.
Anonymized Low costWhat sets it apart
Instruction-following edits from Alibaba that change what you ask for and leave the rest alone, with clean in-image text editing. The line began as an open 20B model, the 3.0 versions are closed, and 2.1 is a 7B model whose Turbo weights are open for research only.
- Changes what you ask, leaves the rest
- Clean text edits in Chinese and English
- Early versions open, Apache 2.0; 2.1 research licence
Recommended for: change one thing and keep everything else ; fix or replace the text on an image.
Strengths and limits
Where it shines
- Changes what you ask, leaves the rest alone
- Edits text in an image cleanly
- About 2 to 2.6 credits an edit
Think twice when
- Anonymized, not private: Alibaba Cloud's policy applies
- 2 Pro is now dearer (about 5 credits) than 3 Pro
- Up to 3 images per edit in Sub Rosa, 5 MB each
- Qwen Image 3, 3 Pro and 2.1: 1K or 2K output
Use cases
Change the text on a flyer
Change the date on this flyer to 'March 14' and keep the same font, color and position.
Text editing inside images is a stated strength of the line.
Swap one object
Replace the laptop on the desk with a stack of three books, same lighting and shadows.
It changes only what you name.
Against its rivals
- Seedream (edit)Seedream preserves lighting and color tone across edits; Qwen is stronger on text.#8 of 26 on Artificial Analysis Image Editing Arena (1,107 Elo)
- GPT Image (edit)GPT Image edits rank higher in the arena; Qwen edits are cheaper.#1 of 26 on Artificial Analysis Image Editing Arena (1,182 Elo)
- FireRed Image EditFireRed is open and private and strong on portraits; Qwen 3 is closed but ranks higher.
Version history
Each version, dated, with what it changed over the one before.
Hollow points are earlier versions no longer offered in Sub Rosa.
See the numbers
| Version | Released | Elo |
|---|---|---|
| Qwen-Image 2.0 (edit) | 2026-03-03 | 962 |
| Qwen-Image 2.0 Pro (edit) | 2026-04-22 | 1,018 |
| Qwen-Image 3.0 and 3.0 Pro (edit) | 2026-07-21 | 1,075 |
| Qwen-Image 2.1 Pro (edit) | 2026-10-02 | 1,069 |
| Qwen-Image 2.1 Turbo (edit) | 2026-10-09 | 1,063 |
- Qwen-Image-EditEarlier, not in Sub Rosa
- Semantic and appearance editing plus bilingual text editing on the 20B Qwen-Image base
- Qwen-Image-Edit-2511Earlier, not in Sub Rosa
- Better consistency: less image drift, better character identity, built-in community LoRAs
- Qwen-Image 2.0 (edit)
- Edit model of the second generation, listed on the leaderboard on 2026-03-03
- Qwen-Image 2.0 Pro (edit)
- Pro tier of the second generation
- Qwen-Image 3.0 and 3.0 Pro (edit)
- Third generation, closed and cloud-only
- Qwen-Image 2.1 (edit)Earlier, not in Sub Rosa
- Unified 7B text-to-image and editing model with native transparent (RGBA) output, up to 10 reference images and local edits by circle, annotation or mask
- Open weights on Hugging Face under the Qwen Research License (non-commercial), a break from the earlier Apache 2.0 releases
- Qwen-Image 2.1 Pro (edit)
- Hosted Pro tier through Qwen Cloud and Alibaba Cloud Model Studio, up to 2048 by 2048 pixels, aspect ratios from 1:8 to 8:1
- Qwen-Image 2.1 Turbo (edit)Latest
- Accelerated checkpoint that renders in 8 denoising steps instead of 40, same 7B architecture
Benchmarks
How to read these scores: what a benchmark tells you.
Artificial Analysis Image Editing Arena: #12 of 26 models measured in the catalog. The same blind duel, for edits: two results of the same instruction on the same photo, and people pick the better edit. An Elo rating: higher is preferred more often. 100 points apart means the higher-rated model wins about 64% of duels.
- 1. GPT Image 2.5 SunburstGPT Image (edit)1,182
- 2. GPT Image 2.5 FlareGPT Image (edit)1,162
- 3. Nano Banana 2.1Nano Banana (edit)1,152
- 4. GPT Image 2GPT Image (edit)1,122
- 5. Muse ImageMuse Image (edit)1,118
- 6. Nano Banana 2Nano Banana (edit)1,108
- 7. Grok Imagine 2.0Grok Imagine (edit)1,107
- 8. Seedream V5 ProSeedream (edit)1,107
- 9. GPT Image 1.5GPT Image (edit)1,102
- 10. Nano Banana ProNano Banana (edit)1,099
- 12. Qwen Image 3 ProQwen Image (edit)1,075
- 15. Qwen Image 2.1 ProQwen Image (edit)1,069
- 17. Qwen Image 2.1 TurboQwen Image (edit)1,063
See the numbers
| Model | Elo |
|---|---|
| 1. GPT Image 2.5 Sunburst | 1,182 |
| 2. GPT Image 2.5 Flare | 1,162 |
| 3. Nano Banana 2.1 | 1,152 |
| 4. GPT Image 2 | 1,122 |
| 5. Muse Image | 1,118 |
| 6. Nano Banana 2 | 1,108 |
| 7. Grok Imagine 2.0 | 1,107 |
| 8. Seedream V5 Pro | 1,107 |
| 9. GPT Image 1.5 | 1,102 |
| 10. Nano Banana Pro | 1,099 |
| 12. Qwen Image 3 Pro | 1,075 |
| 15. Qwen Image 2.1 Pro | 1,069 |
| 17. Qwen Image 2.1 Turbo | 1,063 |
All 6 models in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Qwen Image 2qwen-image-2-edit | Edit | Anonymized | $0.0225 per edit |
Qwen Image 2 Proqwen-image-2-pro-edit | Edit | Anonymized | $0.045 per edit |
Qwen Image 3qwen-image-3-edit | Edit | Anonymized | $0.018 per edit |
Qwen Image 3 Proqwen-image-3-pro-edit | Edit | Anonymized | $0.0225 per edit |
Qwen Image 2.1 Proqwen-image-2-1-pro-edit | Edit | Anonymized | $0.0225 per edit |
Qwen Image 2.1 Turboqwen-image-2-1-turbo-edit | Edit | Anonymized | $0.009 per edit |
Specifications
| Specification | Qwen-Image 3 edit | Qwen-Image 2 edit | Qwen-Image 2.1 Turbo edit | Qwen-Image 2.1 Pro edit |
|---|---|---|---|---|
| Released | July 21, 2026 | March 3, 2026 | October 9, 2026 | October 2, 2026 |
| Parameters | Not disclosed | Not disclosed | 7B | Not disclosed |
| Weights | · | · | Open, Qwen Research License (non-commercial) | · |
| Accepts | text, images | text, images | text, images | text, images |
| Produces | images | images | images | images |
| Resolutions | 1K, 2K | · | 1K, 2K | 1K, 2K |
| Price | $0.018 per edit | $0.0225 per edit | $0.009 per edit | $0.0225 per edit |
| Privacy | Anonymized | Anonymized | Anonymized | Anonymized |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.
Show the 11 sources
- huggingface.co/Qwen/Qwen-Image-Edit
- huggingface.co/Qwen/Qwen-Image-Edit-2511
- drweb.de/qwen-image-3-0-ohne-gewichte-benchmarks
- artificialanalysis.ai/text-to-image/arena/leaderboard-image
- webkul.com/blog/qwen-image-2
- llm-stats.com/blog/research/qwen-image-3-0-launch
- huggingface.co/Qwen/Qwen-Image-2.1-Turbo
- huggingface.co/Qwen/Qwen-Image-2.1
- orcarouter.ai/blog/qwen-image-2-1
- llmreference.com/model/qwen-image-2.1-pro
- marktechpost.com/2026/10/09/alibaba-qwen-releases-qwen-image-2-1-turbo-an-8-step-7b-image-model