Seedance 2.5 vendor terms mapped to Sume request fields

ByteDance and BytePlus describe Seedance 2.5 as multimodal reference-to-video with first and last frame, up to 50 references and 4 to 30 second clips. On Sume's Video Router the same ideas become reference_image_urls, reference_video_urls, reference_audio_urls, image_url, end_image_url and duration, with lower caps: 9 images, 3 videos, 3 audio and 12 in total. This table maps the words to the fields.

The translation table

Vendor wording is from the pages read on 2026-10-04. Sume fields and limits are from the Video Router docs and request schema.

Where the vendor claims come from

The 50-reference figure and the 30, 10 and 10 split are on BytePlus's explainer (BytePlus: What is Seedance 2.5). The 4 to 30 second range, 24 FPS and 480p, 720p, 1080p come from the product page (BytePlus: Dreamina Seedance 2.5). Volcano Engine's page repeats the 50 references and the 4 to 30 second range in Chinese (Volcano Engine: Seedance 2.5).

Two shapes, not one

In the vendor's marketing, first and last frame and references sound like features of one request. Sume models them as two request shapes: image-to-video with image_url and end_image_url, or reference-to-video with reference_*_urls. Mixing them returns 400. The Video Router docs describe each capability; choose the shape that matches the job.

{
  "model": "seedance-2.5",
  "prompt": "The sunrise brightens over the harbor.",
  "image_url": "https://example.com/dawn-start.png",
  "end_image_url": "https://example.com/dawn-end.png",
  "resolution": "720p",
  "duration": 8,
  "mode": "async"
}

Read the catalog, not a blog

Limits move. Call GET /v1/video-router/models and read the capabilities for seedance-2.5 before you hard-code numbers, as the Video Router docs advise. The Video generation docs show the same catalog through /v1/videos with supported_input_references and supported_durations.