Wan 3.0's 20 reference assets vs Omni's 10 images and 3 clips

Alibaba's Wan 3.0 repository lists up to 20 reference assets, including documents, webpages, text and images, while the Sume catalog row for Gemini Omni Flash 1.1 takes up to 10 reference images and up to 3 reference clips of 3 seconds each. If your brief is a pile of documents, Wan's vendor spec is built for it; if your brief is a few photos and a short clip, Omni is enough and has a 4K tier.

The Wan numbers below are the vendor's, from its repository (read 2026-10-05). The Omni numbers are from the Sume Video Router page. Sume's catalog also lists wan-3.0, at 2 to 30 seconds and 480p, 720p or 1080p; read its row from the catalog for its exact reference envelope on Sume rather than assuming the vendor figure carries over.

Side by side

What an Omni reference request looks like

List the media, then address it by index in the prompt. Images and clips each count from zero in the order you list them.

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "<IMAGE_REF_0> holds <IMAGE_REF_1> up to the camera in a bright kitchen, same look as <VIDEO_REF_0>",
  "reference_image_urls": [
    "https://example.com/host.png",
    "https://example.com/product.png"
  ],
  "reference_video_urls": ["https://example.com/look-3s.mp4"],
  "resolution": "720p",
  "duration": 8,
  "aspect_ratio": "9:16"
}

Price context on Sume

Sume lists Wan 3.0 at $0.05, $0.10 and $0.20 per second at 480p, 720p and 1080p, and Omni at $0.03, $0.10, $0.15 and $0.30 for 360p, 720p, 1080p and 4K, both before the 1.25 multiple (Video Router page).

What a reference does and does not do

On the Omni row, a reference guides the look of a new clip; it does not fix a frame. A single reference image with no frame is routed as reference-to-video, and a first or last frame is a different input. If you need the product to appear exactly as in the photo at the start of the clip, use image_url, not a reference. If you need the product to appear anywhere in a scene, a reference is the right call.

The vendor says Wan 3.0 also does instruction- and reference-based video editing, pixel-level box-selection editing and up to 12 sequential images with cohesive style (Alibaba, read 2026-10-05). Those are vendor features; check the Sume catalog row for what Sume exposes.

Test both on one brief

Pick one real brief, for example a 10-second product ad with three photos and one 3-second motion reference. Run it on Omni at 360p ($0.38) and on Wan at 480p ($0.63), compare, then decide which to finish at 1080p ($1.88 and $2.50). Two drafts cost $1.01 together. That is cheaper than a day of debate about spec sheets.

Keep the prompt identical except for the way each model addresses references. Review them blind if you can: have someone else say which is which after they pick.

How to pick

• Brief is mostly documents or text: Wan's vendor spec was designed for that input. Check what Sume's catalog row accepts before you build on it.
• Brief is photos plus a short motion clip, and you want 4K or a draft tier: Omni.
• You want clips longer than 10 seconds in one request: Wan lists 30 seconds, and Sume's row for it goes to 30.
• At 1080p, Omni is $1.88 and Wan $2.50 per 10 s on Sume, so the choice is mostly about inputs and length.