README.md

metadata

title: Nova Streaming Being
sdk: gradio
sdk_version: 5.29.0
app_file: app.py
suggested_hardware: zero-a10g

Nova Streaming Being

This Space is the ZeroGPU renderer-orchestration plane for Nova's conversational digital being.

Adapters

• qwen3_tts_stream: Qwen3-TTS streaming speech target.
• liveavatar_render: Live Avatar target renderer.
• avatarforcing_render: AvatarForcing target renderer, pending public weights.
• fallback_talking_head: Skyreels/OmniAvatar compatible fallback contract.

The Mac mini frontdoor keeps the stable phone UI and business brain. This Space only renders low-latency audio/video segments.

LiveAvatar's published real-time path requires a multi-GPU TPP setup; this Space exposes the stable Nova frontier endpoint, uses Qwen3-TTS streaming when reachable, and records an explicit feasibility verdict for LiveAvatar/AvatarForcing instead of faking 34ms/frame output.