Create AI Avatar · the skill's demo

Three engines, one line.

One digital twin, trained on 2 min 48 s of real takes, saying one line in its own ElevenLabs voice. The same audio on each of HeyGen's engines, so the only difference is the engine.

Judge it by the mouth.

Six moments of each render, nose to chin, and the real footage last.

Four rows of six mouth close-ups: Avatar V, Avatar IV, Avatar III, then the real footage. Avatar V stretches the lips and shows both rows of teeth; IV and III open small and round like the real mouth.
EngineA secondThe mouthEverything else
Avatar V$0.12Lips stretched, both rows of teeth, pinker lipsSmoothed skin
Avatar IV$0.0805Small and round, top teeth now and thenHand gestures
Avatar III$0.01Closest: small openings, natural lip colourThe real head movement

Made with scripts/render.mjs and scripts/mouths.sh. An 8.8 s line cost $1.05 on Avatar V, $0.71 on IV and $0.09 on III.