Product spec sheet to video: Wan 3.0 footage, exact specs via compose
Make a spec-sheet video without a model re-typing your numbers: generate the footage with Wan 3.0, then put your own spec card on screen with Timeline compose.

For a video built from a product spec sheet, let Wan 3.0 make the footage and let your own image carry the numbers. Generate a clip of the product from two or three photos, render the spec table as a still, and put the still on screen over the clip with Sume's Timeline compose. Alibaba's Wan 3.0 README (read 2026-10-05) lists documents as a reference type and text rendering in 12 languages, which makes the model tempting for spec sheets. On Sume the document route is closed (the catalog does not expose file_url), and even where text rendering works, a wrong digit in a spec is a support ticket.
What a spec sheet gives you
A typical sheet has a product photo, a dimensions block, a materials line, a short feature list and a warranty line. Only two of those are good model input: the photo and the feature list as a mood. The dimensions, weights and ratings are facts, and facts should stay in text you control.
- Photos: send as
reference_image_urlstowan-3.0(up to 10 on Sume's catalog entry). - Feature list: compress into a one-sentence action prompt, such as "the lid snaps shut, water beads off the surface".
- Dimensions, weights, ratings: render as a still image from your design tool, not from the model.
- Warranty and legal lines: same, as a still, so the wording is yours.
- Length: 8 to 12 seconds is enough for a product loop plus a spec card.
Compose the spec card over the footage
Sume's Timeline compose docs describe a job that takes one still and one video and returns one MP4 with both on screen. The required fields are operation (stack or overlay), image.url and video.url. The output length comes from the video layer, so the card never makes the clip longer. Both files must already be media.sume.com artifacts or assets of your workspace, so import them first with POST /v1/media-imports.
| Part | Made by | Why |
|---|---|---|
| Product motion and lighting | wan-3.0 from your photos | Generated footage; photos set the product shape |
| Spec numbers and units | Your still image | Exact digits, your fonts |
| Stack or overlay on screen | Timeline compose | One still plus one video, one MP4 |
| Final assembly with other shots | Timeline 1.0 | Ordered slots, transitions, optional audio |
Request
This is the compose call, copied in shape from the docs. Replace the artifact ids with your own after the media import.
Choose overlay when the card should sit on top of the footage, and stack when it should sit beside it. Check the page for the layout options before you rely on a placement.
curl -X POST https://api.sume.com/v1/timeline-1.0/compose \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: spec-compose-001" \
-d '{
"operation": "overlay",
"image": {"url": "https://media.sume.com/artifacts/artf_demo/spec-card.png"},
"video": {"url": "https://media.sume.com/artifacts/artf_demo/product-clip.mp4"}
}'Before it goes live
Open the finished MP4 and compare the card against the sheet number by number. A compose job copies your still, so the digits cannot drift, but a typo in the card image will be copied too. Keep the card file next to the sheet version it came from.
If the sheet changes, you regenerate only the card and re-run compose. The expensive step, the Wan 3.0 footage, stays as it was, which is the main reason to keep facts out of the generated pixels.
Localized versions work the same way. Keep one footage clip and make one card per language; compose is a media job, so each language is a cheap re-run rather than a new generation.
Sources
Related posts
More in Media tools
- Product turntable ad: which Sume video models take two frames
For a product turn, give a video model the front and back photo as first and last frames. Six Sume rows take both; Grok Imagine takes a first frame only.
- Put a promo code on screen in an AI product video with caption cues
Burn a Black Friday code into a silent clip with $0.20 caption cues, key the retry by SKU and code, then pull stills with video-frames to check the text.
- Proof frame for a 1080x1920 export: video frames PNG at 1920
Pull a lossless 1080x1920 still from your vertical export with video frames (format png, max_edge 1920) to check captions and crop before you upload.
- Silent reference clip: reference_ingest's -60 LUFS gate and STT
reference_ingest marks a clip audio.silent at -60 LUFS or below and skips STT, so a silent reference costs no transcript. What to do next: plan new music.
Written by Sume