VIAxVOICE

Face Pipeline v0.1

Drop in a portrait or a short clip. It renders to the ASCII luminance avatar and, in parallel, triangulates the face (468 landmarks) and reads out the expression signals — the data VIAxAI learns its face from. Everything runs in your browser; nothing is uploaded.

Drop an image or video here  ·  or click to choose
JPG / PNG / WEBP · MP4 / WEBM (short clips) — try Athena's portrait
0 frames captured — play a clip while recording

◉ Source · triangulation

no media yet
face tracking: loading model…

▚ ASCII luminance

≋ Expression signals the face's live telemetry — what we learn from

jawOpen over time — the real "mouth envelope". Correlate this with the voice envelope and you have a viseme model, learned from clips.
live viseme →
The mouth shape MediaPipe's blendshapes resolve to, frame by frame. This is the phoneme → blendshape → face bridge — the same path Kokoro audio will drive.

This is the ingestion pipeline (v0.1): image/video → luminance ASCII + MediaPipe Face Landmarker (468-point mesh + 52 blendshapes). Where it grows: batch-ingest clips per voice to learn each persona's face + mouth model, then drive the avatar live from the /synthesize envelope. Voice = identity → face = identity.