Press release to 30-second video: Wan 3.0 prompt and cost on Sume

To turn a press release into a 30 second video on Sume, reduce it to three visual beats, write them as one prompt, and send it to wan-3.0 with duration: 30; at 720p the clip costs $3.75. Sume does not document a document or webpage input for Wan, so the reading and summarizing step happens before the request.

What Sume takes in place of a document

Alibaba's Wan 3.0 repository lists documents and webpages among up to 20 reference assets (read 2026-10-05), and TechNode's launch report is titled around document input. The Sume docs for wan-3.0 list reference images, reference videos and reference audio as URLs, up to 10, 5 and 5. If you paste a release into the prompt, the model sees text, not a parsed document.

What the clip costs

Sume prices Wan 3.0 per output second: provider list x 1.25. The prompt length does not change the price.

A three-beat prompt

Use this skeleton and fill the brackets from the release.

30 second corporate news video, calm pace, no spoken dialogue.
Beat 1 (0-10 s): [setting that shows the problem], [one visual detail].
Beat 2 (10-20 s): [the product or event], [who is using it].
Beat 3 (20-30 s): [the result], end on a clean frame with the text
"[exact headline, 6 words max]" on screen.
Style: [one sentence you reuse in every clip].

Submit it

Send it to POST /v1/videos with model, prompt, duration, resolution and aspect_ratio, and poll the job. Because a 30 second render is a long job, use async or a webhook, not a synchronous wait.

Convert prose into beats

Length is the main choice. A press release has one news item, one quote and one next step, and that fits three beats of about ten seconds each. Longer than 30 seconds, split into a second clip. Shorter than 15 seconds, drop the quote.

Write each beat as what the camera sees, not as what the text says. Models animate pictures; a sentence of corporate prose gives them nothing to draw. "A founder at a standing desk looks at a screen showing a launch dashboard" is usable. "We are thrilled to announce" is not.

Verify every figure and name

Everything you read on screen is a promise to check. If the date, a name or a number appears as text in the video, quote the exact string in the prompt and verify it frame by frame, since Alibaba's repository says only that Wan 3.0 renders long text in 12 languages and gives no accuracy figure (read 2026-10-05). If an exact figure matters, add it with a caption tool in your editor. Sume also documents a standalone caption job, POST /v1/video-captions, which can burn authored overlay cues onto a finished video for $0.20 per job up to 60 seconds.

Format, drafts and total cost

Quote marks matter in the prompt. Anything inside them reads as literal text to render, and anything outside reads as a scene description. Keep literal strings to a headline of about six words and a short call to action.

Pick the aspect ratio for the channel where the video will run. Wan 3.0 on Sume takes adaptive, 16:9, 4:3, 1:1, 3:4 and 9:16, so a newsroom embed is 16:9 and a social post is often 9:16. Make that choice before the first render, because a different ratio is a new paid job.

Draft before you finalize. A 30 second draft at 480p costs $1.875, a quarter of the 1080p price of $7.50. Two drafts and one 1080p final is $3.75 + $7.50 = $11.25, against $15.00 for two 1080p renders. That is the ordinary way to find the wrong beat while it is cheap.

Reviewing the draft

Read the first draft with the press release open beside it. Check three things: whether any figure or name on screen differs from the release, whether the company name is spelled as written, and whether the final five seconds end on a clean frame you can use as a thumbnail. Anything that fails is a prompt problem to fix, then re-run, which at 480p costs $1.875 per 30 second attempt.

Keep the prompt text and the approved clip together in your records. When the next release comes, you can reuse the style sentence and replace only the facts, which keeps a series of announcements visually consistent.

Sources: Wan 3.0 repository and TechNode, both read 2026-10-05; Sume fields from the Video Router docs.