Chinese technology firm ByteDance unveiled Seedance 2.5, the newest iteration of its AI video generation model, at a Beijing conference last month, according to a report from The Information. A separate weekend report indicates the system could be released as early as this week.
AI video generation has advanced considerably since its inception, with each new version delivering increasingly realistic visuals. The earlier, unconvincing clips that once circulated online are now a distant memory.
Today, watermarking for AI-generated videos is essential to help identify deepfakes and other synthetic or false content.
The updated model lets users supply up to 50 reference assets—images, videos, or audio—up from 12 in its predecessor, Seedance 2.0. Expanding reference inputs gives creators greater command over the production process. It can also produce 30-second 4K videos from a single prompt.
ByteDance has regularly shipped leading AI video models, competing with OpenAI’s now-discontinued Sora and Google’s Veo 3. The company, formerly the majority owner of TikTok, is expected to launch the new model in China next month per the report.
AI-focused outlet Testing Catalog reports Seedance 2.5 may arrive on July 9, though the claim is unconfirmed. Still, several platforms have already announced the model is imminent. An official Seedance 2.5 page is live on the Dreamina site, and CapCut posted on July 4 about the upcoming release.
The launch may draw scrutiny. Seedance 2.0’s US debut was delayed earlier this year after Hollywood raised concerns over copyrighted material allegedly used in training. If the new model shows major gains, it could face similar pushback without clearer legal and copyright safeguards.
ByteDance did not immediately respond to a request for comment.