MODEL READY — OMNIPRISM v2 / 256×256
01 — INPUT IMAGEINPUT
Drop image hereClick to browse — PNG, JPG, WEBP
02 — DISASSEMBLED OUTPUTOUTPUT
AWAITING
INPUT IMAGE
INPUT IMAGE
00:00:00system DISASM engine initialised.
02
HOW IT WORKS
UPLOAD INPUT
Provide any assembled object image — CAD render, photograph, technical illustration. Grayscale or colour, any aspect ratio.
NEURAL MAPPING
Gemini Vision describes each part, embeddings guide the slot-attention model, skip connections preserve sharp detail through the decoder.
PARTS LAYOUT
One output image. All disassembled parts composited — variable number of slots activated by the learned gate.