We've added Dreamina Seedance 2.5, ByteDance's next-generation video model, available as Seedance 2.5 (text-to-video and image-to-video) and Seedance 2.5 Reference (reference-to-video).
What's new in Seedance 2.5
- Coherent 30-second clips: The maximum single-request duration doubles from 15 to 30 seconds — tell a complete story without stitching segments.
- Up to 50 reference assets: Reference-to-video now accepts up to 30 images, 10 videos, and 10 audio clips in one request.
- Audio-only references: Drive a generation from reference audio alone — no image or video required.
- First and last frame control: Image-to-video supports a first frame or first + last frames; the output automatically keeps the first frame's aspect ratio.
- Native multilingual speech: Generated dialogue supports 11 languages, including English, Spanish, Portuguese, Japanese, and Korean, with precise lip sync.
- Smart duration: Let the model pick the optimal clip length automatically, or set any whole-second duration from 4 to 30.
Output specifications
Seedance 2.5 generates 480p or 720p video at 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16, with native audio enabled by default.
See the BytePlus ModelArk Seedance 2.5 documentation for full model details.