inclusionAI/Realtime-Venus · Hugging Face

Realtime-Venus-Omni is a 9B audio-visual interaction model, adapted from MiniCPM-o 4.5. It continuously watches and listens, deciding when to respond and generating text and speech on a shared causal timeline. This model supports proactive interaction, semantic interruption handling, and training-free long-video memory. It generates response text and native speech using bundled Token2wav resources and a reference voice.