Separate Image Into Layers With AI

HD (Seedream 5.0 Pro Layerize) picks between 2 and 17 layers itself, names each one and returns the base layer at up to 1536 px. Standard (Qwen Image Layered) splits into the 2–8 layers you choose at about 640 × 640 px and may keep several objects on one layer. HD is the default; Standard suits simple images where a fixed count is enough.

HD costs 7 credits per layer it returns, so an 8-layer result costs 56 credits. When the run starts we hold up to 119 credits, or your whole balance if that is lower (at least 28 are needed), and refund the unused part when it finishes. Standard costs 10 credits per run at any layer count. New accounts start with 0 credits; credit packs and subscriptions are on the pricing page.

With HD, yes: the model names each layer after what it holds, like “Title text” or “Perfume bottle”, and the names carry into the PSD. A short description in the optional prompt box can help. Standard layers get plain numbered names.

HD returns the base layer at up to 1536 px on the long edge; objects come back as cropped layers placed at their original positions. Standard layers come out at about 640 × 640 px. Neither keeps the full resolution of a larger upload, so this is not a route to print-size masks.

JPG, PNG and WebP up to 12 MB, one image per run. Convert HEIC, TIFF or other formats to JPG or PNG first.

Usually one to two minutes. In our tests an 8-layer poster took 85 seconds with HD, and a Standard run took 66 seconds. The panel shows a timer while the model works, and failed runs are refunded automatically.

Usually. The background layer tends to come back with the area behind foreground objects painted in, so moving a separated subject does not leave an empty hole. How convincing the fill looks varies by image.

Not as type. Text arrives as pixels on its own layer, so you can move, hide or recolor it, but changed wording has to be retyped in your editor.

Often, but not perfectly. Glass, smoke, fine hair against busy backgrounds and tiny text are hard, and heavily overlapping objects can end up sharing a layer. Treat the result as a strong first pass and tidy edges in your editor.

Yes. Downloads carry no watermark, and you may use your outputs commercially as long as you own the rights to the image you uploaded.

Your upload and the generated layers are kept in our cloud storage so the preview and downloads work. You can ask support to delete them; the privacy policy has the details.