Wan 3.0 reference limits: how many images, videos and audio clips

Wan 3.0 reference-to-video accepts up to 10 reference images, up to 5 reference video clips and up to 5 reference audio tracks in one request. The video clips together are limited to 15 seconds, and so are the audio tracks. Sume checks the counts when you submit and refuses a request over a cap with 400 unsupported_capability.

The vendor limits are from fal's Wan 3 page and Alibaba Cloud's guide; the Sume checks are from the Video generation docs and current API code. All read 2026-09-29.

What are the exact caps?

What does Sume check on submit?

Only the counts. wan-3.0 is refused with 400 unsupported_capability if a request lists more than 10 image, 5 video or 5 audio references; the error names the input_references field. The 15-second totals and the frame rate are the provider's limits, so keep each request inside them yourself.

Is there also a cap on the whole list?

For Wan 3.0 the per-type caps are the only count check in current code, so the three add up to at most 20. Other models use different rules: for example a general cap of 12 input_references on models without their own per-type caps.

What happens if I send frames and references together?

frame_images wins. The request becomes image-to-video and the input_references are dropped, so send one or the other. Image-to-video vs reference-to-video explains the split.

{
  "model": "wan-3.0",
  "prompt": "The character from the images walks to the window while the song plays",
  "duration": 10,
  "resolution": "720p",
  "input_references": [
    { "type": "image_url", "image_url": { "url": "https://example.com/character.png" } },
    { "type": "audio_url", "audio_url": { "url": "https://example.com/song.mp3" } }
  ]
}