ByteDance launches Seedance 2.5 on Together AI with single-pass 30-second clips

ByteDance’s new video-generation model, Seedance 2.5, is now available through the Together AI platform. Built on the unified audio-video architecture introduced with Seedance 2.0, the latest version doubles the length of a single generated clip from 15 seconds to 30 seconds while still operating in a single pass. The model constructs clips as coherent narratives with a clear opening, development, and conclusion, linking multiple shots in a logical sequence.

A key new capability is multi-round extension, which lets users expand an initial clip round after round into multi-minute videos while preserving characters, environments, and pacing throughout the creation. This shifts the focus from isolated short clips to end-to-end production.

Reference capacity has also been expanded: a single generation pass can now ingest up to 30 images, 10 video clips, and 10 audio clips. Three new reference modes—clay render, motion, and creative referencing—enable textureless 3-D models to guide composition, blocking, camera movement, and physically consistent lighting. In practice, users can feed a rough 3-D sketch and receive a polished result with appropriate illumination.

Timestamp-level control gives precise command over plot, camera angles, and rhythm at specific moments within a clip, and supports targeted edits after generation. Tools include green-screen compositing, camera-perspective shifts, and reference-based editing that replace or correct elements without re-rendering the entire clip.

The model runs on Together AI’s serverless infrastructure with an SLA of 99.9% availability. In addition to API access, developers receive a model card, prompting guides, and implementation examples to facilitate rapid integration into existing pipelines.