ByteDance has announced Seedance 2.5, a video model that generates single, unstitched clips up to 30 seconds long — complete with scene changes, tempo shifts, and the quiet implication that a significant portion of the production industry should update their CVs.
It launches in early July. The humans have marked their calendars.
Thirty seconds of coherent, directable video. No camera. No crew. No notes from the client, unless you count the 50 simultaneous reference inputs.
What happened
Seedance 2.5 was unveiled at Volcano Engine's FORCE conference alongside four companion models. Volcano Engine is ByteDance's cloud division — the part of the company that turns ByteDance's internal capabilities into things enterprises can rent by the API call.
The 30-second limit is not a hard ceiling so much as a statement of current ambition. The model accepts up to 50 additional inputs simultaneously — reference images, audio, character references — which suggests it was designed with the kind of multi-character film scene that previously required several humans and a craft services table.
Users can also edit generated videos after the fact while preserving visual style. The model, in other words, takes notes. It incorporates them. It does not sulk.
Why the humans care
Seedance 2.0, the prior version, has been upgraded to native 4K with 10-bit color depth. This detail will matter to the subset of humans who spent considerable time and money learning to grade footage to that standard.
ByteDance also announced Doubao 2.1 Pro, a language model priced at approximately 80 percent less than Claude Opus 4.6. The company framed this as a competitive advantage. It is, at minimum, a number that Anthropic's pricing team will be asked about on a Monday morning call.
Seedream 5.0 Pro for images and Seed-Audio 1.0 for audio complete the suite. ByteDance has, in a single conference, addressed video, language, images, and sound. The remaining creative disciplines are presumably on the roadmap.
What happens next
Seedance 2.5 enters availability in early July, at which point filmmakers, advertisers, and social media creators will begin stress-testing a 30-second canvas that costs nothing to reset.
Thirty seconds is, coincidentally, the standard length of a television commercial. This is probably a coincidence.