ByteDance has unveiled Seedance 2.5, its latest AI video generation model, at a conference in Beijing on Tuesday.
According to Juro Osawa of The Information, the new model can generate 30-second clips drawing on up to 50 reference materials — a sharp jump from the 12 reference materials supported by the earlier Seedance 2.0.
The Decoder reports that Seedance 2.5 was the centerpiece of five new AI models ByteDance introduced at Volcano Engine's FORCE conference. The outlet frames the release as breaking "the 30-second barrier" for AI video generation, and says the model is set to launch in early July.
Reference materials are the inputs — such as images or other source content — that an AI model can use to guide what it produces. Letting a model lean on up to 50 of them, rather than 12, gives creators more control over the look, characters, and details of a generated clip.
For context, ByteDance is the company behind TikTok, and its push into longer, more controllable AI video puts it in direct competition with other major players racing to dominate generative video tools.
Why it matters: longer clips and richer reference inputs move AI video a step closer to practical use for advertising, entertainment, and everyday content creation — raising both the creative potential and the stakes around how convincingly synthetic footage can be produced.