Mark RadarMARK RADAR
About
EN
Sign in
Event File AI ByteDance

ByteDance Launches SeedRealtime in Doubao App

2 reports · First detected 2026-08-06 · Last active 2026-08-10

AI assistants are moving beyond turn-based text and voice exchanges toward systems that can continuously perceive and respond to their surroundings. SeedRealtime is ByteDance Seed’s attempt to combine listening, viewing, language understanding and speech generation inside one native model, reducing the latency and context loss that can arise when separate components are chained together. The approach puts ByteDance more directly against OpenAI and other developers pursuing real-time multimodal assistants.

ByteDance Seed unveiled SeedRealtime on Aug. 5, 2026, describing it as a native audio-visual full-duplex large language model that processes live audio, video and text streams concurrently. The system can keep watching and listening while it speaks, rather than waiting for rigid conversational turns. ByteDance has deployed the model in the Doubao App, where users can point a smartphone camera at a scene and discuss it with the assistant in real time.

All Coverage

2 original reports

The Backstory

The history behind this event
ByteDance’s Doubao Launches Seed Audio 1.02026-07-22 · 1 reports · similarity 0.82

Generative AI audio is moving beyond standalone speech and music tools toward systems capable of assembling complete soundscapes. ByteDance’s Doubao team has introduced Seed Audio 1.0 for film, video and multimedia production, combining vocals, musical scores and environmental effects within one model. The approach could reduce the need for creators to generate, edit and mix separate audio layers across multiple applications.

Seed Audio 1.0 can produce and integrate voices, background music and ambient sound in a single inference run, creating a unified audio scene, according to the announcement. The model was released for testing through the Volcano Ark experience center at the same time as its unveiling. ByteDance is positioning the system as a cinematic-grade AI audio tool, though commercial pricing and a date for broader availability have not been disclosed.

Mark Radar|MARK RADAR
All times are in Taipei time (GMT+8)