ByteDance launches Seedance 2.5 video generation model
Original: Seedance 2.5
Why This Matters
Seedance 2.5 signals a shift in AI video generation toward full narrative production workflows for professional use.
ByteDance's Seed team officially launched Seedance 2.5 on July 31, 2026. The new video creation model supports up to 30-second single-pass audio-video generation, up to 30 image and 10 video clip references per pass, and timestamp-level editing. It is rolling out on Jimeng AI and Doubao Pro platforms.
ByteDance's Seed team launched Seedance 2.5 on July 31, 2026, describing it as a next-generation video creation model built on the unified multimodal audio-video joint-generation architecture introduced in Seedance 2.0. The key upgrade is single-pass generation extended from 15 to 30 seconds, with support for multi-round extensions that allow users to produce multi-minute content with consistent audiovisual language. The model organizes multiple logically connected shots within a single clip, enabling narrative arcs with setup, development, turning points, and resolution. Multimodal referencing has been significantly expanded: users can now supply up to 30 images, 10 video clips, and 10 audio clips as reference materials in one pass, covering capabilities such as clay render, motion, and creative references. On the editing side, Seedance 2.5 introduces timestamp-level control for targeted audio and video edits, along with enhanced green screen, camera perspective, and reference-based editing features aimed at professional film and advertising workflows. ByteDance demonstrated the model's capabilities through a short creative film produced end-to-end with Seedance 2.5. The model is currently available on Jimeng Web and Doubao Pro under the Video Generation section, with API access planned via BytePlus ModelArk.