Vidu Q4 Preview takes 15 reference images: Sume rows take 9 or 10

Vidu Q4 Preview, launched on 2026-10-07, accepts up to 15 reference images in one reference-to-video request. No Sume video row documents a cap that high. The largest image-reference limits in the Sume catalog are 10 images on Wan 3.0 and on Gemini Omni Flash 1.1, and 9 images on MiniMax H3 and H3 Max. Sume does not list a Vidu model, so a 15-image brief has to be trimmed or split if you run it on Sume.

The Vidu side comes from the launch release, which says the model supports up to 15 image references to define characters, wardrobe, props, products and environments, and from Vidu's reference-to-video API page, which lists 1 to 15 images in png, jpeg, jpg or webp.

Reference limits side by side

The Sume numbers below are the constraints in the Video Router catalog, which is the source behind GET /v1/videos/models. Read supported_input_references and the row's constraints before you build a request, because the envelope differs by model.

How to send a reference job on Sume

Reference images go in input_references on POST /v1/videos. A single reference image with no frame_images is reference-to-video and conditions the whole clip. If you also send frame_images, the request becomes image-to-video and the references are not the mode.

Steps:

• Call GET /v1/videos/models and read supported_input_references for your target row.
• Pick the 9 or 10 images that carry the brief: one per character, then the product, then the location.
• Send them as input_references entries of type image_url with public HTTPS URLs.
• Poll GET /v1/videos/{id} until completed, then fetch unsigned_urls[0].

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-3.0",
    "prompt": "The two founders unbox the bottle in the studio",
    "resolution": "720p",
    "duration": 8,
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/founder-a.png"}},
      {"type": "image_url", "image_url": {"url": "https://example.com/bottle.png"}}
    ]
  }'

Choosing which references to drop

When a brief has fifteen pictures, rank them by how much each one must stay recognisable. A lead character's face and a hero product usually need their own slot. Props, background plates and wardrobe variants can often share one composite image. A contact sheet of three props on a plain background costs one slot instead of three, although the model has to read a busier picture.

If you cannot cut below eleven, split the scene into two jobs that share the same character images and differ in setting. Keep the prompt wording identical for the shared parts so the two clips cut together. That is a workflow choice, not a Sume feature: the platform simply runs each job independently and bills each at its own rate.

What Sume does not do

Sume does not offer a 15-image reference slot on any video row, and it does not list a Vidu Q4 model in the catalog. If your brief needs more than 10 images in one shot, you have two honest options: merge related pictures into a sheet before upload, or split the storyboard into several jobs. The Wan 3.0 onboarding walkthrough prices a 10-image job, and Wan 3.0 at 720p lists at $0.125 a second on Sume.

Vidu's own release notes that pricing, resolutions and feature availability may vary by plan and region, so check its API page before you plan around the 15-image limit.