Employee handbook page to a 20 s training clip: one policy, one scene

A handbook page turns into a good short training clip when the clip shows one behavior and the rule is typed as a caption. On Sume, send a prompt that describes the behavior, optionally with up to 10 reference images such as a floor plan or a neutral workplace photo, to wan-3.0 for 15 to 25 seconds, then burn the policy sentence with video captions cues. Alibaba's Wan 3.0 README (read 2026-10-05) lists documents as reference assets, but Sume has no document field for the model, and a policy is exactly the text a model should not paraphrase.

Choose the page and the behavior

One page, one rule. A page on badge access can become a clip about not holding the door for strangers. A page on lifting can become a clip of a worker bending the knees. Write the rule in one sentence, and then write one sentence of what a viewer would see someone do.

• Rule sentence: copied from the page, exact, for the caption.
• Behavior sentence: a visible action for the prompt, with no text or signage.
• Setting: neutral, generic, with no real company logos or names.
• People: avoid real faces; ask for people seen from behind or at a distance.
• Length: 15 to 25 seconds, so one beat of three or four seconds for each part of the rule.

Build it

The generation request is plain. The caption job is where the policy lives. The captions docs say authored overlay text goes in cues or segments with text, start and end, with no speech-to-text, so the rule appears at the seconds you choose. If you add several short clips for the same course, join them with Timeline 1.0 using ordered slots and fades.

Prompt example

Describe the behavior and the setting, and ask for no readable text.

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: handbook-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A warehouse worker bends their knees and keeps the box close to the body while lifting it onto a shelf, seen from the side. Bright neutral lighting, no readable text, no logos.",
    "reference_image_urls": ["https://example.com/warehouse-ref.jpg"],
    "resolution": "720p",
    "duration": 18,
    "aspect_ratio": "16:9",
    "mode": "async"
  }'

Review before use

A training clip is a statement of company policy, so it needs an owner. Have HR or the policy owner read the captions against the current handbook page, and watch the whole clip for anything that implies something the policy does not say. A model can show a worker doing the wrong thing in the background of a right action, so watch the full frame, not just the subject. Scrub the clip at half speed once, looking only at the edges of the frame, where stray objects and extra people tend to appear. Record the handbook version and the date next to the clip, and regenerate the caption job, not the video, when the wording changes.

Do not use real employees' faces or voices without their written agreement, and do not present a generated clip as footage of your actual workplace. Say in the course notes that the footage is generated. If the policy has legal weight (safety, harassment, data handling), the clip supports the written policy and never replaces it, and the caption should point to the page for the full text.