ByteDance officially released Seedance 2.5 on July 31, moving its video model story from short clips toward controllable production workflows. The Seed team says the model is rolling out on Jimeng AI and Doubao Pro, with API access planned for Volcano Engine's ModelArk. Caixin and Sina Finance also reported the same rollout.
Thirty seconds changes the unit
“The model can organize multiple logically connected shots.”
Seedance 2.5 raises single-generation duration from 15 seconds to 30 seconds and supports multi-round extensions. That matters because 30 seconds is close to the smallest unit of an ad, a classroom explainer or a short-drama scene.
References become production inputs
The model supports up to 30 images, 10 video clips and 10 audio clips in one pass. Dreamina describes this as as many as 50 multimodal references. For brands, that means product shots, character references, music direction and sample footage can sit inside one workflow instead of being scattered across separate tools.
Editing is the harder test
Timestamp-level editing is the most practical feature in the release. Teams need to fix a product placement, a camera move or a line of action without regenerating the full clip. ByteDance also says complex motion and very crowded multi-subject scenes still need improvement.
Caixin reported that Seedance 2.0's overseas call share rose from one-third to one-half by the end of June. That gives 2.5 a user base, not just a demo stage. The next checks are API timing, pricing, concurrency limits and third-party tests of multi-reference and timestamp editing.
Sources: ByteDance Seed official release, Caixin, CocoLoop, Sina Finance; Seedance project page and Dreamina verified 30-second generation, 30-image/10-video/10-audio references, timestamp editing, Jimeng AI/Doubao Pro availability and Volcano Engine ModelArk API scope.