Seedance 2.5: longer AI video and more creative control
ByteDance announced Seedance 2.5 at FORCE 2026. Here is what is known about 30-second clips, multimodal references, availability, and the claims that still need testing.
The name is often misspelled as “Seadance”, but ByteDance officially calls its video generation model Seedance. Following Seedance 2.0, ByteDance announced a successor at the Volcano Engine FORCE conference on 23 June 2026: Seedance 2.5.
From 15 to 30 seconds
The most striking announced improvement is a maximum generation length of 30 seconds in one continuous clip. Seedance 2.0 targeted up to 15 seconds. If that longer generation remains consistent in practice, creators will need to stitch fewer fragments together and can direct action, camera movement, and timing across a longer scene.
More multimodal references
Seedance 2.0 can combine text, images, audio, and video. Seedance 2.5 is reported to support up to 50 multimodal references, which could include characters, locations, style frames, audio, and video fragments. The goal is greater control and continuity within a single generation, although independent use still needs to show how well the model follows a large reference set.
Availability: announced, but still sparsely documented
Reporting from FORCE 2026 says Seedance 2.5 first entered a global enterprise beta, with a public rollout targeted for July 2026. As of 13 July, a complete official 2.5 model card is still absent from ByteDance Seed’s public model page. Pricing, API access, resolutions, regions, and final limits should therefore be treated as provisional.
What does this mean for creators?
Thirty seconds may sound short, but it is a meaningful step for AI video. A complete advertising shot, product demonstration, or short narrative scene can more often be constructed as one unit. That may reduce editing work and makes consistency of characters, movement, lighting, and sound more important than the quality of a single spectacular frame.
The real test follows release
An announcement is not an independent quality test. We particularly want to see how Seedance 2.5 handles complex motion, spoken audio, character continuity, precise camera direction, and long prompts. Copyright, consent for faces and voices, and the availability of effective safeguards also remain essential questions for professional use.
Sources: ByteDance on Seedance 2.0, public ByteDance Seed models, report from the FORCE announcement, and Volcano Engine on the FORCE 2026 video workflow.
All hands on deck?
Tell us what you're building. We'll tell you what we can ship by Friday.