Wan 3.0 reference video requirements: 15 seconds and 16 fps

For wan-3.0 on Sume, reference videos are limited to 5 clips, at most 15 seconds in total, each at 16 frames per second or higher. Alibaba's launch note only says Wan 3.0 can combine up to 5 videos for reference; the 15-second total and the 16 fps floor are Sume's catalog constraints.

Wan facts are from Alibaba Cloud's launch note; Sume facts are from the API catalog, Video inspect and Video trim, read 2026-09-30.

What does the Wan 3.0 launch note say about references?

It says Wan 3.0 can generate up to 30 seconds at up to 1080P, accepts image, text, audio, video, document and web page input, and can combine up to 10 images, 5 videos and 5 audios for reference. It states no per-clip duration or frame-rate rule.

What are the Sume limits?

The wan-3.0 catalog constraint reads: reference_image_urls ≤10; reference_video_urls ≤5 (≤15s total, ≥16fps); reference_audio_urls ≤5 (≤15s total).

How do I check a clip before submitting?

Video inspect returns probe facts for one Sume-hosted media.sume.com clip, with stills and optional transcript. The docs describe the probe object without listing its fields, so confirm what it reports for duration and frame rate. Add up durations across your reference clips and compare to 15 seconds and 16 fps. Import the clip first if it is not already hosted.

What if my clips are too long?

Cut them with Video trim, which returns a new MP4 holding only [start, end) and leaves the source untouched. The range rules are in trim range conflicts. Then pass the trimmed URLs as reference_video_urls. For the full image, video and audio table see Wan 3 reference limits.