Dreamina Launches Seedance 2.5 With 30-Second Clips and Maya and Blender Plugins

Dreamina has launched Seedance 2.5, which it calls a new cinematic video model, in an announcement on X. The company positions the release around native long-form generation and direct plugins for two core 3D applications.

Every capability, quality, and availability figure below comes from Dreamina. The company has not published independent benchmarks, and none of the claims have been verified by third parties.

We covered the teaser when ByteDance first previewed Seedance 2.5's 30-second generation alongside a 4K upgrade to Seedance 2.0. This launch attaches a feature set to that preview.

Native 30-second clips, plus a three-minute long-form mode

Dreamina says Seedance 2.5 generates 30-second clips natively, without stitching shorter segments together. A separate long-video mode extends output to three minutes while holding consistency across the runtime, according to the company.

Runtime has been the main constraint on generative video. A native 30-second clip, if it holds up in practice, covers a meaningful share of real shot lengths, and a three-minute consistent mode would push well past what single-clip tools currently target.

The model ships as plugins inside Maya and Blender

The release includes Dreamina plugins for Autodesk Maya and Blender, which the company frames as a move toward film-grade pipelines. The plugins put the model inside the DCC applications where 3D artists already work, rather than routing them through a separate web tool.

We previously covered AI capabilities arriving inside Nuke, Blender, and Maya through Foundry and Griptape's agent integration. Dreamina's plugins point the same way: generative video as a native step in an existing 3D workflow.

Up to 50 references, including 3D white models and green screen

Seedance 2.5 accepts up to 50 multimodal references per generation, according to Dreamina. The company lists 3D white models (untextured geometry) and green-screen footage among the supported reference types.

That reference count is a control claim. Feeding untextured 3D geometry as a reference would let an artist lock composition, blocking, and camera before the model fills in detail, closer to a previs-to-final pipeline than a single text prompt.

Dreamina also lists two direction controls:

• Interactive in-frame editing. Adjusting elements within a generated frame directly.
• Timestamp control to one second. Specifying what happens at a given second of the clip.

Regional rollout with no US date

Dreamina says Seedance 2.5 is available to subscribers in Southeast Asia, the Middle East, Africa, Europe, and South America. The announcement did not name a United States release or a broader global date.

That staging tracks with how ByteDance has released Seedance before. Seedance 2.0 shipped through several ByteDance platforms, including Dreamina Web and the Doubao app.

Seedance has also drawn IP scrutiny from major studios, which shaped how the earlier model reached users.

Longer runtimes and DCC plugins, but no benchmarks and no US window

Seedance 2.5's claims sit at the high end of what current video models offer: longer runtimes, dense multimodal conditioning, and native plugins for the tools artists already run. None are independently verified, and the announcement leaves out a US launch window.

For 3D and virtual production teams, the Maya and Blender plugins are the detail worth tracking. A generation model that reads untextured geometry and green-screen plates as references, inside the applications where the work already happens, is a more concrete production hook than another standalone text-to-video demo.