Byte quickly follows MiniMax H3 with Seedance 2.5: 30 seconds per generation, up to 50 pieces of reference material per fill

According to ByteBeat Monitor, the ByteDance Seed team has released the video model Seedance 2.5. It continues the architecture of joint input of text, images, videos, and audio, increasing the single-generation duration from 15 seconds to 30 seconds, and supporting further extension of the video.

The model can arrange multiple shots and complete plots within 30 seconds. Users can also repeatedly extend based on the existing results, striving to maintain consistency in characters, scenes, sound, and narrative pace, finally piecing together a video of several minutes.

Seedance 2.5 can input up to 30 images, 10 video clips, and 10 audio clips at a time.

The new model also supports controlling and modifying the video by timestamp. Users can specify what actions or camera shots appear in certain seconds, and can also adjust only the characters, sound, or camera movement without the need to redo the entire segment.

Seedance 2.5 is gradually being launched in the Dream AI and Bean Professional Editions, and the API will soon be integrated into the Volcano Ark.