ByteDance has launched a new AI model, SeeDance 2.5, which allows you to create a 30-second video with a single prompt, use 50 files as references, and edit videos using timecodes.
ByteDance has launched a new AI model, SeeDance 2.5, which allows you to create a 30-second video with a single prompt, use 50 files as references, and edit videos using timecodes.
TikTok parent company ByteDance says it has launched SeeDance 2.5, a new AI model that can generate up to 30 seconds of video with sound at a time. Most of its competitors currently create videos that are 8-15 seconds long per prompt.
The previous model, Seedance 2.0, created videos up to 15 seconds long. The model is said to be able to combine several logically related scenes and build a plot with a beginning, development, and resolution.
The company also presented several video results of the model's work. In one of them, the singer first prepares for the performance in the dressing room, is allowed to leave, gets up, starts singing, and then walks down the corridor backstage, where she meets the dancers, and goes on stage with them. This entire video was created for a single prompt, and not from hand-glued fragments. The finished video can also be continued, preserving the characters, locations, and general style.
“Other AI models can’t do this. Plus, there’s the ability to download up to 50 references to preserve the style, characters, and locations! And in general, this is the best model in the world for dynamic videos. But at the same time, it’s the most expensive in the world (one 30-second video clip in 720p will cost about $7,” notes generative AI consultant Alexey Minakov.
Seedance 2.5 also has more precise timecode editing. The prompt also allows you to specify what should happen at specific seconds. However, it is noted that the model still needs improvement, particularly in terms of generating complex movements and scenes with multiple character interactions.
Seedance 2.5. has already appeared in the Chinese services Jimeng AI and Doubao Pro. Later, the model will become available through BytePlus ModelArk, which will allow integrating the model into third-party applications or services
The new model comes amid ByteDance's conflicts with American film studios and film industry representatives. In March, the company suspended the global launch of SeeDance 2.0 amid claims from major film studios and streaming platforms of copyright infringement. In particular, ByteDance received letters from Disney and Paramount Skydance demanding that it stop unauthorized use of copyrighted materials.
