Continue a Wan 3.0 clip: trim the tail, send it as a reference video

To continue a scene past one clip with Wan 3.0 on Sume, cut the last two to three seconds of the finished clip with video trim and send that piece as reference_video_urls in the next job, with a prompt that says what happens next. Sume's Video generation docs list Wan 3.0 among the models that accept video and audio references, and the catalog constraints for wan-3.0 allow up to 5 reference videos with at most 15 seconds in total, each at 16 fps or more. The vendor, per Alibaba's Wan 3.0 README (read 2026-10-05), also lists reference-based video editing, which is a different feature from this. What follows is a continuation, not an edit.

When to continue instead of lengthening

A single request already goes up to 30 seconds on Sume, so first ask whether you need a continuation at all. Reasons to continue: the scene must run longer than 30 seconds, you liked the first half and want to change the second, or a single long request drifted. The continuation keeps the look of the last seconds as visual guidance, but it does not guarantee a pixel-exact join, so plan a cut or a short dissolve at the seam.

Steps

First, trim the tail. Video trim takes a media.sume.com clip, a start and one of end or duration, and returns a new MP4 of that range. For a 20 second clip, start at 17 with duration 3.

curl -X POST https://api.sume.com/v1/video-trim \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: tail-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/scene-a.mp4",
    "start": 17,
    "duration": 3
  }'

Then send it as a reference

The trimmed file is a new artifact. Use its URL as an entry in reference_video_urls of the next wan-3.0 request. The request needs a URL the provider can fetch over https, so if the artifact URL is not reachable from outside, host the file at a public https address. Describe the continuation in the prompt, starting with where the previous shot ended, and keep the camera and lighting words the same as before.

Joining the parts

When the second clip is done, join the two with Timeline 1.0. Start the second slot where the trimmed tail began, not where the first clip ended, if you want to overlap the repeated seconds; or cut the repeated seconds off the front of the second clip with another trim. Check the seam with two stills, one just before and one just after, before you approve the result. If the character or product changed between the parts, reduce the reference video to two seconds and add reference images of the subject. Keep a log of which tail went with which continuation, because a long scene built from four parts is hard to debug without it.

Remember the budget of reference video seconds: the 15 second total is shared by every reference video in one request, so a three second tail leaves room for other clips such as a style or motion reference. Every part is its own paid generation, so plan the number of parts before you start.