ByteDance launches SeedRealtime 1.0 end-to-end realtime audio-video model API

火山引擎 · wechat · 2026-11-19

Volcano Engine launched the API for SeedRealtime 1.0, a Doubao end-to-end realtime audio-video model that unifies sound, vision, timing and expression in one model, replacing cascaded ASR+VLM+TTS pipelines. It features continuous environment awareness, proactive speech, multi-person visual grounding and noise robustness, demonstrated across six real-world scenarios; dialogue pacing issues halved vs cascaded models.

Original post →

More from Multimodal

Multimodal channel →