According to CoinMeta, ByteDance showcased the video generation model Seedance 2.5 at the Volcano Engine FORCE conference on June 23, with a formal release expected to take place in early July. The main upgrade of this model is that the duration of a single native video can now be doubled to 30 seconds, and it supports importing up to 50 full-modal reference materials at once, aiming to solve the issue of characters and product details deforming during video generation. In addition, Seedance 2.5 also supports fine-tuning of specific parts of the video; users can specify certain actions to occur at particular seconds and which shots to switch between, enhancing the flexibility of video production.