RND in facial animation

What the talk is about

Andrey Sibiryakov talks about R&D in facial animation, drawing on two Metro games and the mistakes he is trying to fix. First, the number of emotion scans grew from 20–30 to 150, with emphasis on their combinations (lips, phonemes, mouth openness); mesh density was tripled and normal maps were added for wrinkles. Cleaning the inside of the lips took five months by hand, but 4D-scan software cut it to two days. Second, he argues for full-performance mocap with actors, instead of recording audio and animation separately, and for neural-network lip sync. Third, lighting, eyes, casting and voice recording can ruin a scene. He urges people to share knowledge openly. Useful for animators, technical artists and developers working on characters.