Ten ad variants from one avatar take: trim, caption, cost vs 10 new

If your ad variants differ in cut, length or caption style, render the avatar once and make the variants with video trim, captions and timeline. On Sume that is about $6.88 for ten variants of one 15-second Plus take, against $38.75 for ten fresh renders with captions. The saving disappears if the variants need a different spoken line, because that takes a new render.

All figures are Sume's published prices on the API pricing page and in the model docs, read on 2026-10-03.

What does each step cost?

Avatar Video is billed per second; the others are flat per job. The video trim and timeline docs describe trim as a flat job and timeline as a per-minute rate. The timeline figure is a reserve based on output minutes, so treat it as a ceiling.

What is the arithmetic for ten variants?

One 15-second Plus take is 15 x $0.245 = $3.675. Each variant then takes a trim ($0.02), a caption job ($0.20) and, if you assemble it with an intro or music, a timeline render of at most $0.10 because a clip under a minute rounds up to one minute. That is at most $0.32 per variant.

Which variants can you really make from one take?

Trim and timeline change what the viewer sees and hears in order; they cannot change what the avatar says. So the cheap route covers:

• Different start and end points, such as a 6-second and a 12-second cut of the same line.
• Different caption styles. The caption job takes a style field, and you can run the same video through several.
• Different intros, outros or music placed by timeline.
• A silent or captions-only version for feeds that start muted.

Which variants need a new render?

A new hook sentence, a different language, a new presenter or a different product image needs a new Avatar Video job, and the renders are where the money goes. The nine-item hook-by-format test in Q4 creative test matrix shows the spend ceiling of that route.

A practical split: spend the render budget on hooks, because the first seconds decide whether anyone watches, and use trims and captions for everything after.

What do you lose with the cheap route?

Variants from one take share a face, a voice and a performance, so a test across them measures cuts and captions, not creative breadth. If one take has a flaw, such as a bad blink or an odd mouth shape, all ten inherit it. Check the take before you fan out; video inspect returns a probe and stills, billed only at its Modal compute.

Also keep each trim's source inside the doc limits: the trim page gives a source up to 1,800 seconds and an output up to 900 seconds, far above any ad.