Start from a photo. The page reads the face on your device, makes a face-blendshape GLB from it, and drives a base VRM's expression blendshapes from that GLB's morphs. Optionally it moves the VRM's face shape, hair shape and colors toward the photo. The skeleton, skin weights and existing blendshapes are not changed, and the hairstyle can be re-laid as waves, curls or coils or switched to a cut preset (no new hair is added).
Stays on this device
Your photo, the face GLB made from it, and everything derived from them are processed in this page and are never sent to any server. The face model runs locally from files served with this page (nothing from a CDN). The page makes GET requests only, and the whole thing works with the network off once the page has loaded.
Approximate, not a likeness guarantee
The binding makes the VRM's expressions drivable from the face source. The shape and color matching below is a rough, measured nudge (eye size and spacing, mouth, jaw, profile depth; skin, hair, eye and brow color); an anime base will still look like an anime character.
Face photo
A clear, front-facing face. The photo stays on this device.
Video only (no microphone), nothing is recorded or uploaded, and the camera is switched off as soon as the selfie is taken. Face the camera, eyes open, even light.
On-device face model: loading…
Download the face GLB made from your photoAlready have a face-blendshape GLB? Use it instead of a photo
Base VRM
Use your own VRM instead (a local file or a URL)
Optional and approximate; everything is applied to the VRM's own geometry and materials.
Re-lays each lock as a wave, curl or coil on the VRM's own hair cards; a cut preset replaces the photo's silhouette; braids, buns and ponytails are not possible.
Pulls hair that the lift has raised off the skull back toward it, so it does not look like a wig; 0 keeps the full lift.
The hairstyle is the VRM's own hair cards, scaled, shifted and optionally re-laid; the texture guessed from the photo is a rough estimate you can override. The body and the skeleton are not changed, and the photo's texture is not painted onto the avatar. With a face GLB instead of a photo only the skin tone applies (there are no measurements or hair/eye/brow colors in it).
Bald binds the VRM again without its hair, so the preview and the download are the same file.
The expression is mapped through the bind map (face morph to VRM expression). The preview is drawn on this device from the bytes in this page.
(no requests)