non-DCT datamosh: held Voronoi cells advect donor frames through current motion
donor: self
no clip loaded
⤓
Drop Host Clip
Drag video here or drop 2 clips for Host + Donor
–/–
graft: each held region keeps the current motion field applied to donor content — flowing forward wearing the wrong skin. Regions are Voronoi cells with noised edges, not macroblocks. At bloom onset the region is seeded from the donor at t + offset; positive = future bleeds backward. Donor can be a second clip (load donor) — the scotoma fills with foreign tissue. Offline: render once, scrub instantly, export WebM.
bloom curve
drag to paint · time →
auto-placement (motion-aware)
⤓
Drop Donor Clip
0% = pure precursor, 100% = pure trailing wake
graft skin
graft flow
scintillating frontier
how long cells scintillate before decaying into interior mush
fringe channel splitting offset
luminance boost on the advancing edge
liquid feedback blur inside older regions
macular distortion
displacement amplitude in pixels
wave frequency (density of ripples)
temporal animation speed of the distortion waves
how many Voronoi regions carve the frame
warps cell boundaries off-grid — membranes, not tiles
how hard held content smears along the flow
where the skin is seeded from at bloom onset. positive = future bleeds backward.