Exclusive: Early 30-second AI videos generated by Seedance 2.5

ByteDance is closing in on the release of Seedance 2.5, the next step in its video model line and the first to generate native 30-second clips in a single pass. The launch was originally penciled in around July 9 but slipped, reportedly amid ongoing talks with rights holders, and the API release is now targeted for July 16. Given how previous Seedance rollouts unfolded, the model may well surface earlier in some places: references to it have been multiplying across partner apps that bundle Seedance models into their video generation and editing toolkits, and pricing details have already appeared on Jimeng in China.

We have had early access to the model, and the first results are hard to argue with. Every clip we generated was a one-shot, 30-second video, and the consistency of characters, lighting, and pacing across that duration is unlike anything else we have tested. Small details occasionally call for a regeneration, but the output holds together remarkably well. The model, unveiled at ByteDance's FORCE conference in June, also accepts up to 50 reference inputs spanning images, video, and audio, and supports region-level edits that change part of a frame without redoing the whole clip.

The rollout is expected to run through Jimeng and Dreamina first, then CapCut and the Volcano Engine API, putting the model in front of advertisers, short-form creators, and ecommerce teams for whom a 30-second spot in one pass removes the stitching step entirely.

The stakes are as competitive as they are technical. Seedance 2.0 topped independent video arenas earlier this year, even as Hollywood cease-and-desist letters slowed its global rollout, and ByteDance has since layered in filters and previewed a copyright-licensing platform. Google's Gemini Omni came close in video editing, but Omni Flash tops out at around 10 seconds per clip, so a coherent one-shot half-minute would reset the gap once again. Mid-July should tell.

Source