Identity is the hard problem in AI characters — and the most satisfying one to solve. Start with training a character LoRA (the identity foundation everything else stands on), then add a synced mouth and a 3D body.
Our presenter, her news-anchor self, and every character voice in our videos share infrastructure: one local LLM, swappable voices, swappable faces. The architecture that makes characters cheap.
The complete recipe behind our presenter’s locked identity — dataset rules, local AI Toolkit vs cloud Civitai training, the strength dials that matter, and the four rules we paid to learn.
The full blueprint for a fully-local AI companion — speech in, animated talking character out — running on one consumer GPU. This is the map; every stage links to a hands-on guide.
MuseTalk generates lip-synced video faster than real time on a 5090 — but getting it to BUILD on Blackwell is dependency hell. Here’s the exact recipe that works.
From a single character image to a fully rigged, animated 3D model you can pose, dress and drive in the browser — an autonomous cloud pipeline with zero local GPU time.