ByteDance releases Seedance multimodal AI video generation models

ByteDance has released Seedance 2.0 and 2.5, representing a significant evolution in AI video generation. Moving away from prompt-only models, the new architecture utilizes a Dual Branch Diffusion Transformer to process visual and audio signals in parallel. This allows for multimodal references, where creators can upload combinations of images, video clips, and audio files to produce consistent outputs.

Seedance 2.5 introduces enhanced capabilities, including the ability to generate audio-video sequences of up to 30 seconds in a single pass. The model supports extensive reference inputs, such as up to 30 images and 10 video or audio clips, to maintain character and environmental consistency. This technology is positioned to serve e-commerce brands, independent filmmakers, and social media creators.

Various platforms now offer access to the Seedance 2.5 model, with different providers specializing in specific workflows such as 4K flagship access, storyboard-driven prompts, or vertical short-form output.