Upload one flat JPG, PNG or WebP and get it back as 2 to 8 separate transparent layers, typically the subject, the background, and shadows or objects. Preview each one, then download a layered PSD or a ZIP of PNG files.
Original
Layer 1Click or drag an image here
Sign in to upload your image
Layer count
Free accounts can use up to 4 layers; subscribers get up to 8.
How it works
A cutout keeps one subject and throws the rest away. Here the whole picture survives: an AI model writes each major part to a separate transparent PNG, and the separated parts line up again when stacked. Photos, posters, product shots, illustrations and AI-generated images all work.
You set how many layers to ask for, from 2 to 8, and the model decides where the boundaries fall. It usually gives the main subject its own layer, keeps the background as a separate plate, and separates out shadows or smaller objects when the count allows.
Results appear next to the original, with a thumbnail strip for clicking through the stack, so you see what was separated before exporting. The preview shows the same files you download, none of them watermarked.
Take a ZIP of numbered PNGs when each part should stay a separate file, or one layered PSD that opens in Photoshop, Photopea, GIMP or Affinity with the stack in order and the background at the bottom.
Sign in, upload one JPG, PNG or WebP of up to 12 MB, pick a layer count and start the run. One to five minutes later you get separate transparent RGBA PNGs that stack up to recreate the original picture, ready to export as a layered PSD or a PNG ZIP. Each run costs 10 credits.
Layer stack
This example started as one flat Hydro Sprint sports-bottle illustration, run at 6 layers; one of its output layers holds the bottle and a checkered strip on an otherwise transparent canvas. Distinct objects with clear edges separate most cleanly, and crowded scenes benefit from a higher count.
Sign in and add a JPG, PNG or WebP of up to 12 MB. A subject in front of a background, or a product beside a few props, gives the model the most to work with.
Choose between 2 and 8 layers: free accounts go up to 4, subscribers up to 8. The run costs 10 credits at any count, and a failed run is refunded automatically.
Click through every result, then download a layered PSD or a PNG ZIP. If the split is too coarse or too busy, change the count and run the image again.
Use cases
One run serves very different jobs. These four come up most often.
Separate the headline, the main image and the backdrop so a layout can be resized or rearranged. Where text lands on its own layer it is pixels, not live type, so plan to retype any wording that changes.
Separated foreground, subject and background layers are raw material for parallax scrolls, 2.5D camera moves and short clips. The background usually comes back with the area behind the subject filled in, so sliding the parts apart leaves fewer holes, though results vary.
When a flattened PNG is the only copy left of an old graphic, the separated layers give you parts to rebuild from. Two or three are enough for a subject on a plain background; busier images need more.
With products, badges and backgrounds on separate layers, a new version is faster to assemble in your editor: swap the backdrop, move a badge, or cover an old caption and set the new one by hand.
Best fit
Worth it when parts of one picture need to move or change separately; the wrong choice when the job needs something the model does not produce.
Photos, posters, product shots and illustrations where the subject, background and props are visually separate. The clearer the edges between them, the cleaner each layer comes out.
Output follows the model’s working resolution, not your original’s: a 1024 × 1024 upload returned 640 × 640 layers. For print-size masks or hair-fine edges, work on the full-size original in your editor.
Each run produces one separated stack that exports two ways: a ZIP of numbered PNGs, bottom layer first, or a layered PSD with the background at the bottom and a flattened preview built in.
Layer basics
Upload the picture here and let the model separate it. The AI runs in the cloud, so a browser is all you need to sign in, upload, preview and download. Photoshop only matters if you pick the PSD export, and Photopea, GIMP and Affinity open that file too.
Usually the main subject or subjects, the background, and shadows or smaller objects, each as a separate transparent RGBA PNG on the same canvas. The model decides the exact split, so two pictures can separate very differently at the same count.
Two or three suit one subject on a simple background; four to eight help with crowded scenes, posters and product shots with props. Free accounts can pick up to 4, subscribers up to 8. A run costs 10 credits at any count, so the only trade-off is how finely the picture is separated.
Signing up is free, and free accounts can run up to 4 layers. Each run costs 10 credits whatever the count, failed runs are refunded automatically, and exports carry no watermark. The pricing page lists current credit packs and subscriptions.
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 five minutes, with a timer in the panel while the model works. If a run fails, the 10 credits return to your balance automatically.
They come out at the model’s working resolution rather than your upload’s: a 1024 × 1024 image returned 640 × 640 layers. That suits social graphics, motion tests, mockups and editing references, but not full-resolution or print-size files.
Yes. A cutout keeps one subject and discards the rest. Here nothing is thrown away: the subject, the background and the smaller parts each end up on a separate layer, and stacking them recreates the original picture.
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, and heavily overlapped scenes need the most touch-up.
Yes. Posters, UI screenshots, illustrations and AI-generated images all work. Text arrives as pixels on a layer rather than editable type, so wording you want to change has to be retyped in your editor.
Often, but not perfectly. Hair, fur, glass, smoke and motion blur hide the boundary between separate objects, and small details can land on the wrong layer or be split across two. Treat the result as a strong first pass and tidy edges in your own editor.
No. The site separates, previews and exports; there is no brush, mask or layer editor in the browser. Editing happens in your own software, such as Photoshop, Photopea, GIMP or Affinity for the PSD, or almost any design tool for the PNGs.
Your upload and the generated layers are kept in cloud storage so the preview and downloads work. Only upload pictures you have the right to edit, and read the privacy policy for how stored files are handled.
Sign in, upload one JPG, PNG or WebP of up to 12 MB and choose 2 to 8 layers. Preview every transparent layer, then download a layered PSD or a PNG ZIP with no watermark.
