Which 10 of 15 reference images to keep for Sume Omni reference video

The short answer

If your brief was written for Vidu Q4 Preview's 15 reference images, cut it to 10 for Sume's Gemini Omni Flash 1.1 by keeping every identity image and dropping repeats of location, style and wardrobe. Sume does not list Vidu. Its Omni row takes up to 10 reference images and up to 3 reference clips of at most 3 seconds each, according to the Video Router docs. The Vidu release states the limit of 15 and nothing about how to weight the images.

A worked cut from 15 to 10

Take a short-drama brief with three characters, a product, a location and a look. The first column is a typical 15-image allocation. The second is what fits in Omni's 10 slots. The rule is that anything the model must recognise on a face or a logo stays, and anything that sets mood can be one image.

How Omni addresses the images

Omni refers to references in the prompt as <IMAGE_REF_0>, <IMAGE_REF_1> and so on, zero-based and in list order. That means the order you send them in is part of the prompt. Put the three faces first, so the prompt can say that <IMAGE_REF_0> is the lead and <IMAGE_REF_1> is her brother. If you reorder the list, rewrite the prompt.

Reference clips work the same way as <VIDEO_REF_0>. Each must be 3 seconds or less and you can send up to 3. They are conditioning inputs, not the thing you edit. The edit mode uses a different field, video_url, and cannot be combined with reference fields.

What the cut costs

A 5-second Omni reference clip at 720p is 5 x $0.125 = $0.625. At 1080p it is 5 x $0.1875 = $0.9375. Sume bills per output second by resolution, so ten references cost the same as one. The cost of the cut is not money but coverage: a location seen from only one angle may drift between shots. If it does, move one wardrobe or style image out and a second location image in, then re-run the same prompt.

When 10 is not enough

If the brief truly needs 15 distinct images, split the scene. Generate establishing shots with the location and style images, then generate the character shots with the faces and wardrobe, and cut them together on the timeline. That costs two clips in place of one, but each clip has its own full set of slots.