ByteDance's Seed research team introduced Seedance 2.5, a new video-generation model designed to give creators more control over longer audiovisual sequences and reference-driven production. The company said the model was rolling out through Jimeng AI and Doubao Pro, while access through the BytePlus ModelArk API was planned for a later date.
The release extends the maximum duration of a single generation from 15 seconds to 30 seconds. ByteDance said users can also extend an existing result through multiple rounds, allowing them to assemble videos lasting several minutes while preserving characters, settings and pacing. The company positions that continuity as a way to reduce the manual work of dividing a story into short clips and repairing transitions after they are joined.
Seedance 2.5 builds on the joint audio-and-video architecture introduced with Seedance 2.0. According to ByteDance, the new model is intended to organize multiple related shots inside one generation rather than merely sustaining a single scene. Its launch materials demonstrate sequences with setup, development and resolution, as well as camera moves that pass between subjects and locations.
The model also broadens the amount of reference material a creator can provide. A single request can include as many as 30 images, 10 video clips and 10 audio clips, the company said. Those inputs can guide subjects, motion, visual treatments and sound. ByteDance highlighted clay-render references and more general creative references among the supported controls, arguing that the expanded input set can better communicate ideas spanning several scenes.
Editing is another focus of the update. Seedance 2.5 offers controls tied to timestamps, enabling users to target particular moments in the generated audio or video. The company also listed green-screen work, camera-perspective changes and reference-based editing as improved capabilities, with film and advertising presented as potential professional uses.
ByteDance described gains in image, motion and audio quality, along with smoother transitions and tighter audiovisual synchronization. Those assessments come from the developer's own launch announcement, and the supplied evidence does not include an independent benchmark or comparative evaluation. The practical consistency of multi-minute extensions and the reliability of detailed edits will therefore depend on testing outside the selected demonstrations.
The launch reflects a broader shift in generative video from producing isolated clips toward supporting an end-to-end creative workflow. The bundled audio generation is important to that proposition because continuity involves voices, effects and timing as well as visual subjects. For Seedance 2.5, ByteDance's central claim is not simply that outputs are longer, but that creators can combine generation, references, extensions and local edits while maintaining a common audiovisual language.
