How it works
Lenticular Depth Print creates a static, lossless two-view simulation from server-estimated relative depth and a semantic foreground mask. A 3 × 3 median filter removes isolated depth speckles, then every model-detected subject pixel plus a narrow anti-bleed boundary is source-locked. Only model-cleared opaque background can move. The two views alternate in exact 1-pixel vertical columns; each view samples one coherent full-RGB coordinate bilinearly. Intensity scales the real two-view geometry through per-view displacement instead of opacity-crossfading completed scenes, with full-strength travel capped at the smaller of 1% of the short side or 10 pixels. Displacement fades to zero at the frame border, and samples are rejected when they cross a depth discontinuity, protected subject boundary, or transparency. Working-frame alpha stays byte-exact and the result is a lossless PNG. This is one flat interlaced image, not an animation and not a print-ready lenticular master.

























