Seedance 2.5: 30-Second Video With Audio in a Single Pass

ByteDance has released Seedance 2.5 — a video generator that produces clips up to 30 seconds long, complete with audio, in a single pass. Previous systems, including Google's comparable offering (Gemini Omni Flash), cap out at around 10 seconds. The new model is already available on the Jimeng AI and Doubao Pro platforms; API access via BytePlus ModelArk will open later.

The model accepts up to 30 images, 10 video clips, and 10 audio tracks as reference inputs simultaneously. This enables complex multi-character scenes with varied camera angles while maintaining visual and audio coherence throughout the entire clip. Textures, lighting, and skin detail are all improved over the previous version.

Director Neill Blomkamp (District 9) used the previous version, Seedance 2.0, to create the 13-minute AI film "Nightborne." According to Artificial Analysis, Seedance 2.0 already tops the leaderboard for image-to-video models with built-in audio. The new version extends capabilities in areas where competitors still fall short — clip length and multimodal input.

The line between professional video production and generative AI continues to blur. The technology is moving toward studio-grade tasks — long scenes with synchronized audio — that just a year ago required days of hands-on work from an entire team.