Upload a photo and get it back with smaller copies of the subject nested inside it, lit the way the cat is.
Everything runs in your browser. The photo is never uploaded anywhere, including to us. A side-on portrait against a plain background works best.
The subject is separated from the background by a segmentation model that runs locally in your browser. Each copy is then scaled down by the step you set, offset, dimmed, and drawn behind the one before it, largest in front.
The model is trained on photographs of people, so it does well with real portraits and struggles with illustration, line art and anything stylised. If it eats the hair or a collar, switch the cutout from Loose to Off and nest the whole frame instead. On drawn art that usually looks better anyway.
Loose grows the mask outward before cutting, which keeps more of the edge. Tight follows the model exactly. Off skips segmentation entirely and nests the full rectangle, which is the Droste version.