QWEN-Image-Edit-2511 — Image Edit AI Model
QWEN Image Edit 2511 rewrites a photo from a plain-language instruction — swap a background, delete an object, restyle the frame — or add a reference image to steer it. Sharp results, and slow: 1-3 minutes.
Also runs in Edit Product Photos by Typing · Outfit Flat-Lay Collage
What it does
QWEN Image Edit 2511 rewrites a photograph from a plain-language instruction - swap the background, delete an object, restyle the frame - and will take one reference image alongside the source to steer the result.
It is sharp, and slow at one to three minutes. Its particular strength is holding a subject's identity steady across an edit instead of letting it drift frame to frame. Output goes up to 1664x928, or square up to 1328.
When to use it
- An edit described in plain language, with no mask to paint first
- Keeping the subject recognisably itself across an edit
- A reference image supplied to steer what the edit should look like
- Product and material variants for design and engineering work
- Portrait retouching, restyling and character work
Details
- CategoryImage Edit
- Year2025
- LicenseApache-2.0
Compliance & Provenance
- ProviderOpen-source (Alibaba) · Specialized
- EU AI Act RiskLimited Risk
- Art. 50 TransparencyRequired — AI-generated outputs are marked
Inputs & Outputs
- ImageInputimage
Source image to edit
- TextInputstring
Text instruction describing the desired edit
- ImageInputimageoptional
Optional reference image for style or content guidance
- ImageOutputimage
Edited output image
Parameters
- stepsintegerdefault 20range 5–40
Diffusion steps (5-40)
Number of denoising diffusion steps. Default 20 — balanced. Lower (5-15) faster but rougher edits; higher (30-40) sharper edits at linear time cost. Bump when the result looks unfinished at default.
- seedintegerdefault -1range -1–2147483647
Random seed (-1 = random)
Random seed for the diffusion sampler. -1 (default) means a freshly randomized seed every run. Pass any non-negative integer (0..2147483647) to make the output deterministic and reproducible. Useful when iterating on prompt phrasing while keeping composition stable.
- resolutionstringdefault Match inputoptions: Match input, 512x512 (1:1), 768x768 (1:1), 1024x1024 (1:1), +7 more
Output resolution (W×H)
Output resolution. 'Match input' (default) keeps the shape of the source image at about 1 megapixel — use it for edits, where a different shape makes the model re-compose the scene instead of editing it. Pick a preset to force one: Square '512x512 (1:1)' / '768x768' / '1024x1024' / '1328x1328'. Wide '1664x928 (16:9)' / '1472x1104 (4:3)' / '1440x960 (3:2)'. Tall '928x1664 (9:16)' / '1104x1472 (3:4)' / '960x1440 (2:3)'.
Alternatives in Image Edit
- FireRed-1.1
Natural-language image edit with strong identity consistency. 1-3 min.
- LatentDiffusion (Object Removal)
Mask-based object removal / inpainting. 10-30s.
- Moebius (Object Removal)
Mask-based object removal. Lighter and sharper than LatentDiffusion. 2-3s.
- PowerPaint
One mask, four jobs — remove, fill from a prompt, rebuild a border, or fill a shape.
- QWEN-Inpaint
Prompt-driven mask inpaint — fills the masked area from your prompt, keeps the rest.
- QWEN-Layered
Decomposes an image into editable RGBA layers (foreground objects, background, …). Pair with List-Extract → QWEN-Image-Edit → List-Inject → Layer-Compose to edit a single layer and recompose.