Course syllabus to a 30 s trailer: Wan 3.0 with one beat per week

A six-week course syllabus fits a 30 second trailer if each week gets one beat of about five seconds. On Sume, wan-3.0 accepts 2 to 30 seconds, so the whole trailer can be one request, with course art as reference_image_urls. If you want to revise week four without paying to regenerate the rest, make six separate clips and join them with Timeline 1.0. The vendor side, from Alibaba's Wan 3.0 README (read 2026-10-05): native 30 second generation, up to 20 reference assets, and automatic scene splitting.

Turn the syllabus into beats

A syllabus lists topics. A trailer needs pictures. For each week, write one visual verb: a thing that happens on screen that a viewer connects to the topic. Week 1, "foundations", can be a hand placing the first block. Week 3, "data", can be rows of light filling a grid. Keep it concrete, and avoid asking for text such as week numbers in the picture.

Check that no two weeks share a verb. If weeks two and five are both "building something", the trailer will blur them. A useful test is to read the six verbs aloud with the topic names removed: a viewer should still be able to tell the beats apart. Also decide the audience before you write, because a trailer for working professionals and one for first-year students need different pacing even with the same syllabus.

• List the six weeks with one noun and one verb each.
• Choose one or two images that set the look: a course cover, a palette swatch, an instructor-free illustration.
• Write the prompt as six numbered beats of five seconds each.
• End with a still hold for the enrollment line, added later as a caption.
• Decide up front: one 30 s request, or six 5 s requests joined by Timeline.

One request or six

The Sume docs give both routes. One request is simplest and keeps a continuous look. Six requests give per-week control and a cheaper retry if one beat is wrong. Timeline 1.0 takes ordered video[] slots and returns one MP4; the Timeline docs set its duration per slot, accept transitions after the first slot (fade, dissolve, wipes and slides), and cap a transition at 1 second.

Prompt and request shape

Reuse the same reference images in every clip so the six beats share a look. The request below is the single-job version. If you pick the six-clip route, keep the prompt the same except for the beat, and keep the order of the reference images unchanged so the look does not move between jobs.

Each job is asynchronous: you submit, receive a job, and poll or use a callback. Send a different Idempotency-Key for each clip, and the same key only when you are retrying the same clip. Reusing one key across six different prompts will be read as a repeat of the first request.

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: syllabus-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A six-week course trailer in six five-second beats: placing blocks, rows of light filling a grid, a bridge being drawn, gears meshing, a map unfolding, a finished model turning. Calm palette, no text.",
    "reference_image_urls": ["https://example.com/course-art-1.png", "https://example.com/course-art-2.png"],
    "resolution": "720p",
    "duration": 30,
    "aspect_ratio": "16:9",
    "mode": "async"
  }'

Finishing

Add the course name and enrollment date as captions rather than generated text, using the cues input described in video captions. The enrollment date is exactly the kind of fact a model should not draw: a wrong digit there costs real sign-ups.

If the trailer needs sound, the Wan 3.0 clip can carry generated audio, or you can mute it and lay your own audio spine under the clips in Timeline 1.0. Decide that before you render, since the spine length sets the total length of the timeline output.