ByteDance's Seedance 2.5 Leaks: Native 30-Second AI Video Generation, API Expected July 16

ByteDance's upcoming AI video model Seedance 2.5 has surfaced through exclusive early footage shared by TestingCatalog. The model generates native 30-second video clips in a single pass — no stitching multiple short segments together, but a single inference producing a coherent half-minute video.

First unveiled at ByteDance's FORCE conference in June, the launch was originally penciled in for around July 9 but slipped amid ongoing talks with rights holders. The API release is now targeted for July 16. However, references to Seedance 2.5 have already appeared across partner apps that bundle Seedance models into their video generation toolkits, and pricing details have surfaced on ByteDance's Jimeng platform in China.

In TestingCatalog's hands-on testing, every generated clip was a one-shot 30-second video with remarkable consistency in characters, lighting, and pacing across the full duration. The model accepts up to 50 reference inputs spanning images, video, and audio, and supports region-level editing that modifies part of a frame without regenerating the entire clip.

The Jimeng platform has also revealed an "Ultra-Long Video (Beta)" mode powered by Seedance 2.5, capable of generating videos up to 180 seconds long, further expanding the model's capability envelope.

Seedance 2.5 is expected to roll out first through Jimeng and Dreamina, then CapCut and the Volcano Engine API, targeting advertisers, short-form creators, and ecommerce teams who need coherent 30-second spots without post-production stitching.

Seedance 2.0 topped independent video benchmarks earlier this year, despite Hollywood cease-and-desist letters that slowed its global rollout. ByteDance has since added content filters and previewed a copyright licensing platform. Google's Gemini Omni is competitive in video editing, but Omni Flash tops out at around 10 seconds per clip — Seedance 2.5's native 30-second output could reset the competitive gap.