Hot on the heels of MiniMax's H3, ByteDance's Seed team has released Seedance 2.5, its latest video-generation model. Like its predecessor, Seedance 2.5 takes text, images, video, and audio as combined inputs — but now it can generate a single clip up to 30 seconds long, double the previous 15-second limit, and it can keep going from there.
The model is designed to storyboard multiple shots and a complete narrative arc within those 30 seconds. Users can also iteratively extend an existing result, with the model doing its best to keep characters, scenes, sound, and pacing consistent — letting you stitch together a multi-minute video piece by piece.
As for inputs, Seedance 2.5 accepts up to 30 images, 10 video clips, and 10 audio tracks in one go.