doubao-seedance-2-5-260628) now sits alongside 2.0. All four task types — text-to-video, first frame, first/last frame, and multimodal — plus aspect ratio, resolution and duration are picked directly on the page, so you can run through the output quality and parameter combinations before deciding how to wire it into your own service. Set duration explicitly: SeeDance 2.5 supports 4–30 seconds, and the smart setting typically resolves to about 10 seconds, roughly twice the cost of 5 seconds.
The SeeDance 2.5 / 2.0 generator in iCover AI: model, aspect ratio, resolution and duration all selectable on the page
asset:// ID back, then insert it from the “asset” tab on each media slot. The request body shrinks to a few dozen bytes, the create-task call returns immediately, content checks move up to ingest time, and asset IDs stay reusable across tasks so a character keeps its look. The reference-image slot on the test page caps at 12 as a demo limit; the API itself supports up to 30 images plus 10 video and 10 audio references, so call it directly per the integration docs for batch work.
For the latency breakdown and migration steps behind the asset-first flow, see Asset-First: Faster, More Reliable Image and Video Inputs.
← Back to Live Updates · 📚 Monthly archive