ByteDance releases Seedance 2.5, generating 30 seconds of audio and video in one pass

ByteDance has released Seedance 2.5, a video-generation model that produces up to 30 seconds of synchronized audio and video in a single pass. The company’s Seed research team detailed the model in a July 31, 2026 blog post.

Most AI video tools generate silent clips that need sound added separately, or cap continuous output at a few seconds. Seedance 2.5 targets longer, sound-complete takes, with what ByteDance calls multi-round extensions to stretch scenes past the 30-second mark.

The model accepts up to 30 images, 10 video clips and 10 audio clips as reference material in a single input, ByteDance said. It adds timestamp-level control for targeted audio and video edits, plus clay-render referencing, green-screen editing and camera-perspective control.

Seedance 2.5 is rolling out now inside Jimeng AI, ByteDance’s creative app, and the Pro tier of Doubao, its consumer assistant. Developer access through the Ark platform on BytePlus, ByteDance’s cloud arm also known as Volcano Engine, is expected soon.

The performance claims come from ByteDance and have not been independently verified. The company published sample outputs but no third-party benchmark or side-by-side test against rival systems such as OpenAI’s Sora or Google’s Veo, and quality on 30-second single-pass generation is hard to judge from curated demos.

Days after the quiet release, Seedance 2.5 drew renewed attention among developers, a sign the audio-video pairing is landing in a market where synchronized sound has been the missing piece. ByteDance has not disclosed pricing for the coming API.