Seedance 2.5: A Single-Pass 30 Second Video with Audio

Recently, ByteDance has released its one-shot video synthesis model — Seedance 2.5, which can generate up to 30 seconds of videos with audios. Compared with other models like Google’s Gemini Omni Flash, Seedance 2.5 is able to generate longer videos (up to 10 seconds). Currently, Seedance 2.5 is integrated into Jimeng AI and Doubao Pro, and it will be exposed via BytePlus ModelArk’s API shortly.

Compared with Seedance 2.0, Seedance 2.5 supports more reference images(up to 30), videos(up to 10) and audios(up to 10). With these rich references, complex scenes containing multiple characters and different perspectives can be easily generated. Meanwhile, Seedance 2.5 also keeps the overall consistency of the whole video and audio. Besides, Seedance 2.5 has better texture details, lighting conditions, and skin tones than Seedance 2.0.

Director Neill Blomkamp (District 9) used Seedance 2.0 to produce an AI film named “Nightborne”. The length of “Nightborne” is 13 minutes. According to Artificial Analysis, Seedance 2.0 is the state-of-the-art image-to-video model with sound. Seedance 2.5 inherits all the advantages of Seedance 2.0 and introduces some new functions like longer video length and multi-modal inputs.

With the development of technology, video generation is getting closer and closer to generative AI. Soon, we can do professional video editing just like a video studio. Then, we can edit long scenes with synchronized audios, which may cost us several days when working with a team.