OmniFaceRig: Fully Automatic Inner-Mouth-Aware Face Rigging Across Diverse 3D Character Topologies

OmniFaceRig is aimed squarely at one of the most tedious parts of character production: getting a usable facial rig out of a static mesh. The paper describes a fully automatic pipeline that takes a surface-only character and produces an inner-mouth-aware FACS rig, complete with procedurally fitted teeth, gums, tongue, and up to 155 blendshapes.

What makes it notable for game teams is the scope of the automation. The system is designed to work without manual landmarks, per-character template edits, or user-provided oral-cavity setup. It combines VLM+CV-based riggability checks, multi-model face parsing, dense keypoint registration, procedural inner-mouth construction, and collision-aware blendshape transfer. In practice, that means less technical artist time spent on repetitive setup and fewer hand-authored fixes for mouth interiors that break during animation.

The paper says it supports humans, humanoids, and a range of animal topologies, including long-muzzled and short-muzzled characters such as dogs, wolves, foxes, cats, bears, rabbits, and tigers. For non-human characters, it picks topology-specific face and inner-mouth templates and tries to reduce teeth-face intersections automatically, which is the kind of problem that usually becomes a bespoke rigging headache.

The authors also released Omni-Bench, a dataset of 1,000 biped 3D characters with FACS facial blendshapes and inner-mouth geometry. For developers, the practical implication is less about a magic one-click rig and more about a plausible path to scaling facial setup for generated or heavily variant...