Wan 3.0 reference videos: 15 seconds total and a 16 fps floor on Sume

Alibaba advertises Wan 3.0 as accepting up to 10 images, 5 videos and 5 audio files as references. Sume's wan-3.0 catalog row enforces the same counts and adds two limits that the headline hides: reference videos total at most 15 seconds and must be at least 16 fps, and reference audio totals at most 15 seconds. Plan the length of your reference clips before you plan the count.

This matters because five 10-second reference clips would pass a count check and still be refused.

Vendor claim versus the Sume row

The count limits match. The duration and frame-rate limits come from the Sume catalog constraints for the row, which is what the validator applies at submit time.

Budget the 15 seconds

Fifteen seconds is a pool shared by every reference video in the request, not a per-file allowance. Three clips of 5 seconds each use it all. If you have a long source clip, cut the part you need first with video trim rather than sending the whole file.

• One 12-second motion reference leaves 3 seconds for a second clip.
• Five reference audio files still share one 15-second pool.
• A reference video under 16 fps fails the row's constraint even if it is short enough.

What the request looks like

Reference media goes in the same Video Router body as the prompt. URLs must be public HTTPS addresses, so a file on your laptop has to be uploaded first. The example sends one image, one clip and one audio file.

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: wan3-refs-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "The mascot from the image waves, matching the clip and the voice",
    "reference_image_urls": ["https://example.com/mascot.png"],
    "reference_video_urls": ["https://example.com/motion-6s.mp4"],
    "reference_audio_urls": ["https://example.com/voice-8s.mp3"],
    "resolution": "720p",
    "duration": 8,
    "mode": "async"
  }'

When to pick another row

If you need longer reference clips, the other rows differ. The Video Router page lists each model's limits, and a Seedance or H3 row has its own counts. Check the live catalog, then test one short request before a batch.

Source for the vendor numbers: Alibaba Cloud's launch note, read 2026-10-04.