ByteDance has launched SeedRealtime, a native audio-video full-duplex large model, according to a post on the company’s Seed WeChat account.

ByteDance said the model integrates audio, video, and text in a unified architecture, which allows real-time interaction across continuous multimodal information streams.

SeedRealtime appears to come from ByteDance’s core AI research effort rather than a specific product team.

Seed describes itself as a foundational research unit established in 2023. Its work spans large language models, speech, vision, world models, and AI infrastructure.

It also says its models are powering over 50 real-world applications.

Any commercial rollout in China would likely face deep-synthesis rules that require labeling of AI-generated content and impose obligations including real-name verification, content review, logging, and security assessments for services with broad public reach.

🔗 Source: ByteDance