Spreadsheet to AI video: Wan 3.0 xls input vs chart images on Sume

You cannot send an .xlsx file to wan-3.0 on Sume. Alibaba's page lists xls, ppt, pdf and other document formats as Wan 3.0 inputs, but that is the vendor's own endpoint on Alibaba Cloud Model Studio. Sume's /v1/videos docs list image_url, video_url and audio_url references and first/last frames, and no document type. The practical route is to render each chart to a PNG and hand the images to Sume as references or frames.

What each side says

The workflow that does work

Export each chart or table from the sheet as an image. Keep it at the aspect ratio of the clip you plan to make. Then pick one of two modes.

For a morph between two states, such as last quarter's bar chart and this quarter's, send the first as a frame_images entry with frame_type first_frame and the second as last_frame. Sume treats this as image-to-video. A 10 s clip at 720p costs 10 x $0.10 x 1.25 = $1.25.

For a longer piece that uses several charts, send up to 10 images as input_references. Sume's docs describe references as visual guidance rather than exact frames, so numbers on a chart can drift. If you send both frame_images and input_references, frame_images controls the mode and the references are not used as references, so choose one.

Request for the two-chart morph

The image URLs must be reachable by the provider, so host the exported PNGs somewhere public or on Sume's media storage.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: q3-vs-q4-chart-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "The bars of the first chart grow and reshape into the second chart. Clean white background, no extra text.",
    "duration": 10,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "frame_images": [
      {"type": "image_url", "image_url": {"url": "https://example.com/chart-q3.png"}, "frame_type": "first_frame"},
      {"type": "image_url", "image_url": {"url": "https://example.com/chart-q4.png"}, "frame_type": "last_frame"}
    ]
  }'

Check the numbers on screen

Alibaba's FAQ names on-screen text rendering accuracy as a current limitation. For a finance chart that matters: read every axis label and value in the output against the sheet before you publish. A safer pattern is to generate the motion at 480p ($0.0625 per second) and add real labels later in a timeline, using the exact figures from the sheet rather than model-drawn text.

Sume's timeline and media tools work on Sume-hosted media, so the overlay step stays inside the same workspace. Check the timeline docs for the exact clip and overlay rules before you plan it, because they take media.sume.com artifacts rather than arbitrary URLs.

Why not send the sheet itself

It is tempting to screenshot the whole spreadsheet and send it as one image. Resist it: a screenshot of 40 rows becomes a texture, and the model will not read cell values from it reliably. One chart per image, with large labels and few series, gives the model a shape to animate and gives you a clean frame to compare against afterwards. Alibaba's own description of the feature is animated, dynamic charts, which is a visual treatment and not a data pipeline.

Limits to plan for

• Ten reference images is the cap for wan-3.0 in Sume's request validation; a 14-tab workbook needs batching.
• One job covers 2 to 30 seconds, so a 30 s explainer is $3.75 at 720p and $7.50 at 1080p.
• On the Video Router's /v1/video-router/generate endpoint, reference audio is rejected unless a reference image or video comes with it.