Wan 3.0 box-selection editing: Sume offers Omni video_url edit instead

You cannot draw a box and edit inside it with wan-3.0 on Sume: the Sume docs list no box, mask or region field for Wan. The edit path Sume documents is gemini-omni-flash-1.1 with a video_url and a text instruction, billed at $0.125 per output second at 720p.

What the vendor says and what Sume documents

Alibaba's Wan 3.0 repository lists "pixel-level box-selection editing" among its editing features, next to instruction-based and reference-based editing (read 2026-10-05). It is a vendor capability on Alibaba's own surfaces. The Sume Wan entry is documented as text-to-video, image-to-video with an optional end frame, and reference-to-video, with audio. Editing is not on that list.

The edit route Sume has

The Video Router docs describe one edit mode, on gemini-omni-flash-1.1. You send video_url (the source clip) and a prompt that gives the instructions. resolution is optional with a default of 720p. You cannot send aspect_ratio or duration, because the source clip sets the output. The source cannot be combined with image_url, end_image_url or reference_*_urls.

Describing a region without a box

No mask is involved, so you describe the region in words. That is a real difference from drawing a box, and the quality of the result depends on how precisely the prompt names the object and its position.

• Name the object and the area: "the bottle on the left of the table", not "the product".
• State the change and the constraint together: "Replace the bottle with an apple. Keep everything else the same." That is the example in the Sume docs.
• Make one change per request, so you can see which instruction caused which difference.
• Keep the source clip short; the edit is billed for the output seconds.

Request

The request shape is from the Video Router docs.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: video-router-edit-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Replace the bottle on the left with an apple. Keep everything else the same.",
    "video_url": "https://example.com/clip.mp4",
    "resolution": "720p",
    "mode": "async"
  }'

What an edit costs

Sume bills Omni Flash 1.1 at the provider list times 1.25 for each output second: the catalog list is $0.03, $0.10, $0.15 and $0.30 a second at 360p, 720p, 1080p and 4K (fal, 2026-08-28). Because the edit length follows the source, the budget is the source length times the rate. The docs add that a duration sent with a video_url is only a reserve hint, with a default of 8 seconds.

Google's version versus Sume's

Google's own post says Gemini Omni 1.1 Flash supports 720p standard, 1080p or 4K premium output, and scene extension up to 40 seconds total in 10 second steps (read 2026-10-05). The Sume row is narrower: 3 to 10 seconds for generation, and edit length set by the source. Use the Sume docs for what you can send here.

Edit or regenerate

If your real need is a different clip, not a changed one, regenerate. A new 10 second Wan 3.0 clip at 720p costs 10 x $0.125 = $1.25, the same per-second rate as an Omni 720p edit, and Wan reaches 30 seconds in one request. If your need is to keep every pixel except one object, an edit is the only route Sume documents, and a manual mask in your own editor is the alternative.

Reduce the ambiguity

A box is a precise instruction because it removes ambiguity about where the change goes. Words reintroduce that ambiguity, so the practical risk is a model changing the wrong object or touching the background. Reduce that risk with a short source clip, a prompt that names unique visual features (a red cap, the third book from the left), and a check pass at 360p before a 720p or 1080p edit. Omni's 360p rate on Sume is $0.0375 a second, so a 10 second draft edit is $0.375.

Plan on one or two re-runs per difficult edit. That is a planning assumption rather than a measured rate. If the third attempt still moves the wrong object, switch to a manual mask in a video editor and use the model only for the creative step.

Which model for which job

Pick Omni when the source clip already exists and must stay. Pick Wan 3.0 when you are making a new clip and can describe the desired result from scratch, up to 30 seconds. Both are priced identically at 720p on Sume, so the difference is control, not money.

Sources: Wan 3.0 repository (read 2026-10-05), Google's Gemini Omni 1.1 Flash post (read 2026-10-05), and the Video Router docs.