Seedance 2.5 nears launch with videos up to three minutes

Beijing, China. ByteDance is preparing to release its Seedance 2.5 video-generation model through Dreamina and CapCut. Dreamina’s official product page already lists the model’s capabilities, but it also labels Seedance 2.5 as “coming soon”. It would therefore be premature to describe the global rollout as complete.

The headline change is duration. Seedance 2.5 is presented as generating continuous clips of up to 30 seconds in standard mode and outputs of up to 180 seconds through an experimental long-video mode. Dreamina also says a single task can use as many as 50 multimodal inputs, including text, images, video, audio, scripts and style references.

What is officially claimed for Seedance 2.5

• single-scene generation of up to 30 seconds;
• a beta long-video mode extending output to 180 seconds;
• up to 50 multimodal references in one project;
• localized editing without regenerating the entire video;
• motion and composition control using video and 3D references.

CapCut’s Seedance 2.5 page emphasizes 30-second scenes, timestamped instructions and camera control. Separate CapCut promotional pages also promise native 4K output. These specifications concern an unreleased product and have not yet been independently tested.

Three minutes may not mean one flawless take

Dreamina’s wording separates 30-second generation in standard mode from the 180-second extension in beta. The public description does not make clear whether the longer workflow produces all three minutes in one model pass or uses an internal continuation and scene-assembly process.

Maintaining character identity, motion physics, lighting and camera logic remains difficult as AI-generated video becomes longer. Support for 50 references could give directors more control, but it also raises the importance of a coherent script and compatible source materials.

When Seedance 2.5 could become available

No exact date has been confirmed. TestingCatalog reports a possible July launch and stresses that a July 9 date remains a rumor. A separate Dreamina page gives roughly mid-July as a target, while access may vary by region, account and product surface.

ByteDance has not publicly detailed Seedance 2.5 pricing, API terms, launch regions or the limits of its long-video beta. Users should therefore check availability directly inside Dreamina or CapCut rather than rely on third-party sites already advertising access before the public release.

Why the model is drawing attention

ByteDance’s previous model already ranks highly in independent comparisons. At the time of checking, Dreamina Seedance 2.0 720p led Artificial Analysis’ text-to-video leaderboard and also ranked first for image-to-video generation. The platform listed an indicative price of $9.07 per minute, although rankings and pricing can change.

Seedance 2.5 shifts the contest from short showcase clips toward advertising, product demonstrations, music videos and compact narratives. Cifrum.kz previously examined why claims about new models still require open testing in its report on Meta’s unreleased Watermelon model.

Another trend is the expansion of AI capabilities inside mainstream consumer products. One recent example is Google Maps’ planned Gemini food-ordering feature. In both cases, practical availability, price and reliability matter as much as the model’s advertised capability.

Sources: the official Dreamina Seedance 2.5 page, CapCut product materials, TestingCatalog and the current Artificial Analysis rankings.

The lead image was created with artificial intelligence for Cifrum.kz as a conceptual editorial illustration. It does not depict the real Seedance 2.5, Dreamina or CapCut interface.