Tsinghua's Bunraku Turns Single Illustrations into Editable Live2D Characters

bdsqlsz · x · 2026-08-18

Researchers from Tsinghua University and others introduced Bunraku, a method that automatically converts a single static illustration into an editable, drivable Live2D character. The process involves splitting the image into an ordered RGBA layer stack and building a triangle mesh for each layer to predict 2D displacements. It requires no per-example tuning or manual editing, producing raw model output.

Related event: Bunraku Turns Single Illustrations into Editable Live2D Characters(2 posts)→

Original post →

More from Multimodal

Multimodal channel →