MACE

Masky Avatar Compute Engine

MACE

Pilot a Masky avatar from your machine — webcam expressions and lipsync on your CUDA GPU, no per-minute cloud render bill.

Get the engine See the pipeline

Inside the Masky ecosystem

Sign on with Masky, pick an avatar, then run face + audio locally. MACE ships approved ARKit blendshapes to mace.masky.ai so Masky can reload known rigs for that avatar later.

  1. 01 Login with Masky — avatar identity & voice
  2. 02 Photo → ACE-ready head (52 ARKit shapes) on your GPU
  3. 03 Review & approve shapes; contribute training to the shared model
  4. 04 Live: webcam drives expression, lipsync audio can be Hume-converted

Realtime, local cost

Capture mic and camera on-device. Heavy inference stays on your NVIDIA GPU. Cloud is for auth, avatar catalog, voice conversion, and shared blendshape memory — not pixel streaming.

  • ARKit / ACE-compatible rigs from a single Masky portrait
  • Human review gate before a rig is marked ready
  • Shared suggester improves as users contribute

HTTP API

Base path https://mace.masky.ai/api/v1. Masky services call this internally when they need stored blendshapes for an avatarId. 

POST /avatars/{avatarId}/rig/suggest
POST /avatars/{avatarId}/rig/contributions
GET  /avatars/{avatarId}/rig
GET  /health

Full contract on GitHub →