ByteDance launches SeedRealtime for real-time audio-video interaction

ByteDance's Seed team has launched SeedRealtime, a native audio-video full-duplex large model that integrates audio, video, and text in a unified architecture for real-time interaction on continuous multimodal streams. The model excels in joint audio-video understanding, resolving homophone ambiguity with context and grasping visual temporal references. It also features proactive interaction, detecting environmental changes and alerting users, plus tool integration. Its interaction rhythm is smooth, sensing conversation state and resisting background noise. End-to-end tests show 50% fewer dialogue rhythm issues compared to cascade models, with higher odds of smooth single conversations. SeedRealtime is now fully available on the Doudao app—update to the latest version, select "Call" in the chat box, and enter video call mode to try it.