Seedance 2.5 reference limits: 30 images on the vendor page, 9 on Sume

ByteDance's Seedance 2.5 page says you can provide up to 30 images, 10 video clips, and 10 audio clips as references in a single pass. Sume's Video Router request schema is stricter: for seedance-2.5 it accepts at most 9 reference_image_urls, 3 reference_video_urls, 3 reference_audio_urls, and 12 reference files in total.

If you plan a build around the vendor number, design for the lower cap on Sume and test it. A request over the cap returns a 400 before any money is reserved.

The two sets of limits side by side

The vendor figure describes the model. The Sume figure describes what the API will accept today. They can change at different times, so read the live GET /v1/video-router/models response before you hard-code either.

Request rules that also apply

Audio references need company. reference_audio_urls requires at least one reference image or video, and the API rejects audio-only input. First and last frame fields cannot be combined with reference_*_urls, so pick image-to-video or reference-to-video for a given job.

All media must be public HTTPS URLs that Sume can fetch. A URL that is not reachable fails the job, so check each one with a HEAD request first.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ref-demo-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "The mug from the first photo on the desk from the second",
    "reference_image_urls": [
      "https://example.com/mug.png",
      "https://example.com/desk.png"
    ],
    "reference_audio_urls": ["https://example.com/room-tone.mp3"],
    "duration": 12,
    "resolution": "720p",
    "mode": "async"
  }'

How to pick your references

Fewer, cleaner references usually beat a pile. Start with the 2-4 images that carry the look you care about, then add a video or audio reference only when motion or sound is the point. Keep a log of which set produced which clip, keyed by your idempotency key.