Dreamina Seedance 2.5 Arrives on BytePlus ModelArk With Watermarks and Face Restrictions

On August 6, 2026, BytePlus announced that Dreamina Seedance 2.5, the model ByteDance Seed had unveiled on July 31, is available to developers and enterprises through its ModelArk platform. This is the API route that the original launch had listed as coming soon.

Details

• Clip length: Up to 30 seconds per clip, aimed at longer stories with smooth transitions and stronger character consistency.
• References: Up to 50 references per clip across images, video, audio and text.
• Editing: Specific timestamps, characters or other elements of a generated video can be adjusted while keeping the scene consistent.
• Target use: BytePlus emphasizes advertising, marketing and e-commerce workflows.
• Access: ModelArk in supported markets globally, the BytePlus console, and third-party platforms named in the post (Higgsfield, Runway, DreamBench). BytePlus is not available in the United States.
• Safeguards: Visible watermarks plus C2PA Content Credentials; harmful-content filters under the GenAI Acceptable Use Policy; protections against unauthorized IP; and restrictions on videos made from images or videos containing real faces. Vetted clients can complete real-person verification and likeness authorization in the ModelArk console. ModelArk also offers a library of more than 10,000 virtual-human assets.
• Model ID and pricing (from the ModelArk pricing page, not the blog post): the page carries a banner saying the official Seedance 2.5 API is live, lists the model ID dreamina-seedance-2-5-260628, and shows $10.70 and $11.70 per million tokens in online inference for the two input types in its table (offline inference is listed as not supported). Check the table itself for which price applies to which input, since we could not confirm the column labels. The blog post gives no price, resolution or frame rate.

What happened next

The API followed the Seed launch by about a week, though we found no separate API press release beyond the BytePlus product, blog and pricing pages. For teams, the practical points are the model ID, token-based pricing that varies with whether video is in the input, and the real-face restrictions, which will shape which reference workflows are allowed. Early third-party pages disagreed on whether the model was in ModelArk; the official pricing page now settles that.