Start with the face
Three MetaHuman characters replace the earlier avatar samples. Their facial rigs include jaw, lip, eyelid and cheek shapes. Normal and roughness maps keep skin detail; hair uses strand coverage instead of a flat colour texture.
ROBOTCANDY EXPERIMENT / DIGITAL ACTORS
A little more human.
A face, a voice, a moment of doubt.
Direct the light. Listen
closely.
A SIX-LINE SCENE
The second shadow
Preparing the first portrait…
WHAT CHANGED / WHAT COMES NEXT
Better faces need
more than better light.
This experiment renders actual 3D meshes on your device. The recordings are prepared; the face, lighting and camera are live.
Light for shape
Large key, fill and rim lights, soft shadows, eye reflections and filmic tone mapping. Switch to Flat to see what the same face loses without directional lighting. This comparison changes the light, not the character.
Let sounds shape the mouth
The written dialogue is aligned to the recordings before deployment. Each face has its own jaw and lip settings. Lip contact takes priority for b, p and m, while vowel shapes flow into each other. Choose Previous version to compare the earlier performance. This is prepared animation, not captured acting.
The remaining ceiling
Browser MetaHumans retain more facial detail than the previous samples. They do not reproduce Unreal's skin scattering, strand hair, wrinkle correctives or captured performance. High-quality baked facial motion is the next improvement; neural Gaussian avatars are promising but need dedicated capture and training.