Slide deck to video: ten slides as Wan 3.0 reference images on Sume

To turn a slide deck into a video on Sume, export up to ten slides as public HTTPS images and send them as reference_image_urls to wan-3.0; a 30 second clip at 720p costs $3.75. Sume does not take a slide file or a PDF for Wan, so the export step is yours.

What fits the request

Alibaba's Wan 3.0 repository describes up to 20 reference assets, including documents and webpages (read 2026-10-05). The Sume schema is narrower: it caps wan-3.0 at 10 reference images, 5 reference videos and 5 reference audio files, and every reference is a URL. A slide deck therefore enters as a set of images.

The workflow

Export the slides you want the video to follow as PNG or JPG files, host them at public HTTPS URLs, and keep them in order. Sume rejects an eleventh image with a validation error naming reference_image_urls.

• Pick at most ten slides. If the deck is longer, make several clips and join them.
• Export each slide as an image and upload it to a public HTTPS location.
• Describe in the prompt what the video should do with them: order, tone, and what text stays readable.
• Submit once at 480p to check the sequence, then render the final.

Request body

The request below uses the documented input_references field of POST /v1/videos. The video-router route takes reference_image_urls for the same purpose.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: deck-video-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A 30 second product walkthrough that follows the slides in order",
    "resolution": "720p",
    "duration": 30,
    "aspect_ratio": "16:9",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/s1.png"}},
      {"type": "image_url", "image_url": {"url": "https://example.com/s2.png"}}
    ]
  }'

What references do

References steer the look; they are not guaranteed frames. The docs say the model uses them as visual guidance, not exact frames, so do not expect each slide to appear for equal time.

Cost

Cost is duration times the Sume rate. The reference count does not change the Wan price in the repo's pricing table.

When a reference is the wrong tool

If you need the actual slide to appear sharp for several seconds, an image-to-video start frame (image_url) on a short clip is a better fit than a reference. For text-heavy slides, keep the resolution at 720p or 1080p. Related limits are listed in the reference limits post.

Prepare the slides

Slides are a poor match for video until you shape them. A deck written for a presenter has dense text, small type and no motion. Before export, do three things: cut each slide to one idea, enlarge the one line of text you need legible, and remove footers and logos you do not want animated. A deck of ten slides becomes ten clean images.

Order matters because references are listed in order in the request. Number the files so the list in your request is the order you want, and say in the prompt what each part should show. The model follows the visual style of the images, so a consistent template across slides gives a more consistent clip.

Pick ratio and length

Aspect ratio comes from the request, not from the slides. Wan 3.0 on Sume takes adaptive, 16:9, 4:3, 1:1, 3:4 and 9:16. A 16:9 deck suits 16:9 output; for a vertical short, export or crop slides to 9:16 first so type is not cut off. Use adaptive if you want the model to choose.

Duration is the other choice. Ten slides in 30 seconds is three seconds per slide, which is quick for text. If the audience must read each slide, make three 10 second clips of three or four slides each. The per-second price is the same, so splitting costs nothing extra; it only adds a join step in your editor.

Where Sume differs from the vendor

Vendor context: Alibaba lists documents and webpages among the reference assets Wan 3.0 accepts (read 2026-10-05). On Sume, the documented Wan inputs are image, video and audio URLs, so we cannot promise direct document upload there. Slides as images is the workflow that matches what Sume documents today. Check the Video Router page before you build around anything beyond that.

Sources: Video generation docs, Video Router docs, and the vendor's Wan 3.0 repository (read 2026-10-05).