Six shots, three takes each: Vidu Q4 $2.84 to $35.10, Sume from $11.25

A 30-second scene of six 5-second shots with three takes each is 90 generated seconds. Cost per kept second at Vidu Q4 and Sume rates is in the table.

5 min readSume
All posts

The short answer

A 30-second finished scene made from six 5-second shots, with three takes per shot, means you generate 6 x 5 x 3 = 90 seconds to keep 30. At Vidu Q4 Preview's 540P list rate that is 90 x $0.045 = $4.05, or $2.835 with the image-to-video discount. At the 4K list rate it is $35.10. Sume does not list Vidu; on Wan 3.0 at 720p the same 90 seconds cost 90 x $0.125 = $11.25. Three takes is a planning number, not a measured rate.

Cost per finished second

The last column divides the total by the 30 seconds you keep, which is the number to quote to a client.

Six 5-second shots, three takes each, 30 seconds kept, as of 2026-10-08 (Vidu: press release, read 2026-10-08; Sume: public catalog, read 2026-10-08)
RowArithmeticTotal for 90 sPer kept second
Vidu Q4 540P, 30% off90 x $0.0315$2.835$0.0945
Vidu Q4 540P, list90 x $0.045$4.05$0.135
Vidu Q4 4K, 30% off90 x $0.273$24.57$0.819
Vidu Q4 4K, list90 x $0.39$35.10$1.17
Sume wan-3.0, 720p90 x $0.125$11.25$0.375
Sume gemini-omni-flash-1.1, 1080p90 x $0.1875$16.875$0.5625
Sume seedance-2, 720p90 x $0.378$34.02$1.134

Why three takes

Video models are not deterministic. Sume's docs say no v1 video model accepts a seed, so you cannot pin a result and vary one thing; a re-run is a new roll. For a face or a hand that must be right, expect to discard some takes. Three is a reasonable start for planning. Replace it with your own ratio after the first twenty clips.

A useful trick is to spend takes at a cheap tier and the final render at an expensive one. Find the framing at 720p, then make the keeper only once at 1080p. On Sume, one extra 1080p take on Wan 3.0 for a 5-second shot is 5 x $0.25 = $1.25.

What changes the number

Longer shots lower the share spent on shot overhead but raise the cost of each discard. Shorter shots, 3 seconds, cut the discard cost but add cuts, and some Sume rows set minimums: Seedance 2.0 accepts 4 to 15 seconds and Omni accepts 3 to 10. Check the row's valid length in the video generation docs before you plan the shot list around it. For the Vidu discount window, see the test-batch planner.

Reading the table

Per kept second, Vidu 540P with the discount is $0.0945 and Wan 3.0 at 720p is $0.375, so the Sume row is 4.0x. At 4K list Vidu is $1.17 per kept second, and Omni at 4K would be 90 x $0.375 / 30 = $1.125. Those two sit close together, which is the main finding: at the top tier the price per second is nearly the same and the choice rests on length limits and look. Seedance 2.0 at 720p is $1.134 per kept second, three times the price of a Wan 720p take, which is worth paying only if the output is visibly better.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume