ByteDance has released Seedance 2.5, an update to its AI video generator that produces clips up to 30 seconds long with dialogue, music and sound effects created in the same pass. The model went live in early August 2026 on the company's Jimeng AI and Doubao Pro apps, with programming access through BytePlus ModelArk to follow, according to The Decoder.
The headline change is length paired with sound. Most rival tools still cap a single generation at a handful of seconds and add audio afterward. Seedance 2.5 builds the picture and the soundtrack together, so speech and background noise line up without manual syncing. ByteDance says the clips can then be extended for longer sequences.
The model also takes far more source material as guidance. Users can feed it up to 30 photos, 10 video clips and 10 audio files in a single request. The company says this helps keep several characters and camera angles consistent across a scene. Other reported gains include sharper textures, lighting and skin detail.
For non-specialists, the point is simple. A 30-second take is closer to a finished scene than the 5-to-15-second snippets typical of current tools. ByteDance showed this off with a short film, "The Missing Pair," made entirely with the model. Director Neill Blomkamp earlier used the previous version, Seedance 2.0, for "Nightborne," a 13-minute AI-generated short.
The length also sets Seedance apart from its closest competitor. Google's Gemini Omni Flash generates clips of roughly 10 seconds, about a third of what ByteDance now offers. On quality, the earlier Seedance 2.0 already ranks first among audio-capable image-to-video models on a public leaderboard run by Artificial Analysis, an independent benchmarking firm. Its position there is based on blind user votes.
Beyond the specifications, the release fits a wider shift. Video generators are moving from silent, short clips toward full audiovisual scenes, and ByteDance brings an unusual edge. It owns TikTok and Douyin, among the largest libraries of short video to train on. That is one reason its models have led on motion and realism.
Two cautions remain. Most of the figures come from ByteDance's own launch materials, and independent tests at scale have not yet arrived. The company has also not published a price or a confirmed maximum resolution. Longer, sound-synced clips lower the cost of making video, which helps small creators and advertisers, but it also makes convincing fake footage cheaper to produce. Whether Seedance 2.5 keeps its lead will depend on outside benchmarks, and on how fast Google and others answer with longer clips of their own.
