QWEN-Image-Layered — Image Edit AI Model
QWEN Layered splits one picture into stacked RGBA layers, background first, so a single object can be edited without disturbing anything behind it. Pull a layer out, edit it, push it back, recompose.
What it does
QWEN Layered splits one flat picture into stacked RGBA layers, background first, so a single object sits on its own transparent layer the way it would in Photoshop or Figma.
That is what makes a one-thing edit possible: pull a layer out with List-Extract, edit it with QWEN-Image-Edit, push it back with List-Inject and recompose with Layer-Compose, and everything you did not touch comes back unchanged. Three, four or eight layers, at 640 or 1024.
When to use it
- Changing one object while the rest of the picture stays untouched
- Moving, resizing or recolouring an object without repainting it
- Editing a text block on its own layer, the artwork left alone
- Deleting an object cleanly, with transparency already behind it
- Decomposing a layer further, by feeding it back in on its own
Details
- CategoryImage Edit
- Year2025
- LicenseApache-2.0
Compliance & Provenance
- ProviderAlibaba (open weights) · Specialized
- EU AI Act RiskLimited Risk
- Art. 50 TransparencyRequired — AI-generated outputs are marked
Inputs & Outputs
- ImageInputimage
Source image to decompose into layers
- TextInputstringoptional
Optional prompt to guide the decomposition (leave empty to auto-decompose)
- LayersOutputlist:image
Decomposed RGBA layers (background first)
Parameters
- layersstringdefault 4options: 3, 4, 8
Number of layers to decompose into.
How many RGBA layers to split the image into: '3', '4' (default) or '8'. More layers separate finer elements but cost more time and can over-segment simple images.
- resolutionstringdefault 640options: 640, 1024
Working resolution / quality (output aspect follows the input).
Working-resolution bucket — sets quality, NOT the exact output size (each layer's W×H is derived from the input image's aspect ratio at this area). '640' (default) is faster; '1024' gives higher detail at ~2.5× the time. Only these two values are valid.
- stepsintegerdefault 50range 10–60
Diffusion steps (10-60)
Number of denoising steps. Default 50 — balanced. Lower (10-30) is faster but rougher layer separation; higher (50-60) is cleaner at linear time cost.
- seedintegerdefault -1range -1–2147483647
Random seed (-1 = random)
Random seed for the diffusion sampler. -1 (default) randomizes every run. Pass any non-negative integer (0..2147483647) to make the decomposition deterministic and reproducible.
- true_cfg_scalefloatdefault 4range 1–10advanced
CFG scale (1-10, 4.0 default).
Classifier-free guidance strength. Higher follows the prompt more strictly; lower stays closer to the source image. 4.0 (default) is balanced.
- negative_promptstringdefault advanced
Negative prompt — what to avoid. Empty for none.
Tags
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-2511
Edits images from a natural-language instruction. High quality, and heavy (1–3 min).
- QWEN-Inpaint
Prompt-driven mask inpaint — fills the masked area from your prompt, keeps the rest.