Where VIAxAI grows a face. Every tool here runs on the one signal VIAxVOICE already emits — the amplitude envelope + facial data — turning voice into a moving face, in ASCII or on a real photograph.
The generative face. Mouths any line via a viseme table learned from a real performance, with an auto emotion layer (reads the sentence) and adjustable ASCII detail. The chosen VIAxAI look.
Open avatar →Ingest a portrait or a clip → ASCII luminance render + 468-point facial-landmark triangulation. Records per-frame mesh + expression and exports the training data. The front end of the learning loop.
Open pipeline →Remap the mesh onto a real image: warp a photo's face to follow a driving clip — mouth, brows, eyes. Eye-frame registration normalizes scale so faces match. The "onto images" path.
Open reenactment →The hybrid: a real face wears the voice. Load a portrait, type a line — the mesh deforms from the viseme signals and the photo speaks it as high-fidelity ASCII. No driving clip. Runs entirely on your device.
Open speaking face →Where the face goes next: real phoneme timing for sharper lips, a face per voice, audio-driven photo (a still speaks with no clip), and generative graphical communication.
See roadmap →These are working experiments, not finished product — expect rough edges (the reenactment especially is compute-heavy and iterating). The through-line is fixed: one facial-data engine, many faces. — Kairos