AvatarGen is a local app for building animatable avatars — runs on macOS,
Windows, and Linux. You generate a body in
Tripo3D from a photo, get an animatable
head from AvatarSDK, drop both into the
app, and out comes a single rigged GLB with 51 ARKit blendshapes — ready to
drop into a Babylon.js / three.js / Unity / WebXR scene with working lip-sync
and facial expressions.
Pick the right neckline. The graft replaces
the head and a short neck, so the body image must be clothed at the
base of the neck. Use a simple round / crew neckline that
covers the chest. Avoid tall stand / mandarin collars (they sit
in the head-removal zone and get stripped) and open / scooped / bare
necklines (the small neck seal helps only narrow join gaps, not a
wide bare neckline). Keep hair short or off the neck, face the camera, no hat or
scarf. Full rules and a ready-to-paste image prompt are in the
guide.
Critical: in Tripo's UI, run
Segmentation → Highbefore clicking Animation.
Tripo's "Animation" step is what generates the rig; once a model is
rigged you cannot add segmentation retroactively. Without
segmentation the graft can't cleanly remove the Tripo head and you'll
see skin-coloured "stain" triangles on the AvatarSDK face.
Correct order: Generate → Segmentation High → Animation → Download GLB.
Get the animatable head from AvatarSDK in
FBX format — not OBJ, GLB, or PLY. Only the
FBX export contains the 51 ARKit blendshapes the pipeline needs for
lip-sync and facial expressions.
Drop both into the Build avatar tab and click Generate. You get
an integrated, animatable GLB with the AvatarSDK head replacing Tripo's
and all clothing preserved.
Optional: use Add hat to overlay a hat GLB with live positioning,
or Edit avatar to manually clean up any residual artifacts by
clicking on them.
What the integrated GLB contains
Tripo body mesh — full clothing, materials, skeleton preserved
AvatarSDK head meshes (AvatarHead, eyelashes, eyeballs, teeth, hair)
parented to the Tripo Head bone
51 ARKit blendshapes on AvatarHead — jawOpen,
mouthSmileLeft, eyeBlinkLeft, etc. — drive these for
facial expression and lip-sync from any source (Azure / ElevenLabs viseme
streams, Rhubarb Lip-Sync, Web Audio amplitude analysis, etc.)
Typical file size 8–14 MB; the user guide shows how to compress with
gltf-transform for WebXR deployment