A full-stack, face tracked, protogen head software package. Aimed at running on a Raspberry Pi 5, this operating system consists of four independent aspects, working together to bring yourself to life! Allowing you to control your face with... your face.

Those being:

• A procedurally generated face, taking in standard facial blendshapes to controls.
• A face tracker.
• An eye tracker.
• And a voice modulator.

A protogen is an anthropomorhic cyborg. Relevant to this project: it has a visor with a digital display for a face.

The face generator, facegen/: blendshapes in over OSC, LED panel frames out. Built and running end to end; see its README for how to run and test it.

• The face is rendered on the GPU. The Pi 5's VideoCore VII is driven through Vulkan (via wgpu), and the eye, mouth and nose are signed distance fields written as WGSL fragment shaders. Faces are shader code plus a TOML file of millimetre parameters; editing either reloads live. The same shaders run unmodified in a browser through WebGPU.
• Every LED panel is a window onto one face-space. All panels are rendered into a single atlas laid out as the LED driver's framebuffer, each carrying its own affine transform, so panel count, size, pitch and mounting are layout data. Two 64x32 panels and a six-panel visor are shipped presets.
• Blendshapes drive shape parameters through a rig. Project Babble's 45 shapes, the ARKit set and the protOS eye and voice channels are one vocabulary; a gain table in the face TOML maps them onto per-side parameters, so expressions can be asymmetric.
• The mouth listens. By default it is a scope: a thin line along the lip carrying the voice spectrum, bass at the snout tip and treble toward the corner, under a travelling sine that slows to a gentle wobble in silence. The 32 bands come over OSC from whatever is analysing the microphone; the band definition lives in protos-audio/ so every producer agrees. A jaw mode with lips and teeth is a TOML switch away.
• A browser harness shows the panels live, with a slider for every input, layout editing and a scene switch. A spinning cube scene turns each panel into a window onto a small 3D world, which is the start of display pieces beyond the face.
• Panels are driven by Adafruit Piomatter through piomatter-sys/, a small binding over its C++ core, behind a cargo feature for the Pi.

See it running, rendered from the built-in curves:

• Captures of the face and the cube on both presets.
• Live demo: the renderer compiled to WebAssembly, rendering with WebGPU in the browser, with every input on a slider.

MIT, see LICENSE. The one exception is piomatter-sys/, a binding to Adafruit Piomatter, which is GPL-2.0-only; a facegen binary built with its piomatter feature is GPL-2.0-only when distributed. The Project Babble tracking model is fetched at setup time under its own non-commercial licence and is never committed.