Generate at 720p, then upscale: $0.134 a second vs Vidu Q4 4K at $0.39

The short answer

If you want higher-resolution output on Sume and can accept an upscaled result in place of a native render, generate at 720p and run Sume's Video Upscale. A 720p row such as Wan 3.0 or Gemini Omni Flash 1.1 costs $0.125 per second, and Video Upscale costs $0.009 per input second, so the pair is $0.134 per second. Vidu Q4 Preview's 4K rate is $0.39 per second. Sume does not list Vidu. The Sume catalog entry for Video Upscale does not state its output size, so test a sample before treating the two as equal.

Per-second and 10-second costs

Video Upscale is priced per input second after Sume's margin, accepts up to 30 seconds, and reserves 5 seconds if you give no duration.

What you are trading

A native 4K render carries detail that the model drew at that size. An upscale enlarges a 720p frame, and it can sharpen edges but it cannot add detail that was never drawn. For a talking head on a phone, the difference may not show. For a product macro with fine text, it will.

The saving is real if the result passes your review: the 10-second 720p-plus-upscale route is 10 x $0.134 = $1.34 against $3.90 for Vidu 4K at list and $3.75 for native Omni 4K. Against Vidu's discounted 4K rate of $0.273, it is 10 x $0.273 = $2.73, still twice as much.

A safe way to test it

Generate one 5-second clip at 720p, which costs 5 x $0.125 = $0.625, and upscale it for 5 x $0.009 = $0.045, so the test is $0.67. Compare it frame by frame with a native 4K Omni clip of the same prompt, which costs 5 x $0.375 = $1.875. If you cannot tell them apart at your delivery size, you save about 64% on the rest of the project. If you can, you have learned that for $2.55. Read the Video Router docs for the request shape.

Limits that affect the maths

Video Upscale accepts at most 30 seconds of input, which costs 30 x $0.009 = $0.27. A longer piece has to be upscaled in sections. The 720p generation step has its own ceiling, so a row such as Omni, which stops at 10 seconds, needs several clips and several upscales for a longer film. Add the cost of joining them if you use the timeline. Plan the length of each section before you generate, since the upscale charge is per input second and a trimmed clip costs less to enlarge.