PhotonTTS (work in progress)

Fast, human-like, zero-shot voice-cloning TTS with singing support. Design: a causal LM backbone at 12.5 Hz over continuous VAE latents, a per-patch flow-matching head distilled to 1-2 steps, and Score Clock conditioning for singing. Tiers: Photon-S (edge/CPU), Photon-M (consumer GPU), Photon-L (24GB+).

Status: scaffolding. No weights yet.