How it works
Paper Depth Diorama estimates relative distance, smooths the depth field, removes small label islands, and divides the prepared fixed frame into up to seven broad paper layers. A semantic mask keeps one reliably detected main person, pet, product, or object on one coherent sheet instead of slicing it at noisy depth thresholds. Each nearer sheet receives a narrow warm cut highlight and casts a soft lower-right shadow only onto a farther receiving sheet; zero-padded offsets cannot wrap a false shadow across the frame. Source luminance is simplified into a warm paper palette, and intensity changes the real palette, shadow offset, blur, opacity, and cut highlight. At its working size, the renderer never moves, copies, spatially resamples, inpaints, or generatively rebuilds source content: every RGB treatment stays at its original coordinate, framing remains fixed, and working-frame alpha stays byte-exact in the lossless PNG. Larger uploads can be proportionally resized before this step: the standard free limit is 800 pixels on the long side, and the depth service defaults to a 2048-pixel long side for remaining larger inputs. If one reliable main subject cannot be isolated, the effect stops safely instead of guessing.
