$5 of avatar video: 27 seconds standard, 20 plus, 9 max

Five dollars buys 27 seconds of Sume Avatar 1.0 talking video at standard (5 / 0.184 = 27.17), 20 seconds at plus (5 / 0.245 = 20.41) or 9 seconds at max (5 / 0.55 = 9.09) (as of 2026-10-09). If you also have to create the avatar at $0.95, the $4.05 left buys 22, 16 or 7 seconds.

Whole seconds from a fixed budget

Sume prices per second of video, and the docs accept 4 to 60 seconds, so the practical question is how many seconds a budget can pay for. Round down, because a script that runs a second over its estimate raises the price.

What fits in those seconds

A 27-second standard video is one hook, one benefit and one call to action at a conversational pace. A 9-second max video is a single greeting. The 4-second minimum in the docs means max's 9 seconds allows two shots at most. Most people draft at standard and keep max for the final cut.

The preview flow lets you test the first frame without paying for the render. After you approve it, call generate-video and pick the tier; the stills are tier-independent.

Splitting $5 across two clips

Two clips are often better than one long one. At standard, a 13-second clip is 13 x 0.184 = $2.392, and two of them are $4.784, leaving $0.216. At plus, two 10-second clips are $4.90, leaving $0.10. Each clip needs at least 4 seconds of estimated length, so the minimum viable pair is 8 seconds.

Do the split when you want two versions of the same message, for example 9:16 and 16:9, or a hook test. Each version is one job, so keep them under the budget with a margin. One avatar handle serves both.

Keep a reserve

Jobs reserve their estimated price at submit and refund it if the job fails. If a balance is close to a job's price, the API can answer 402 insufficient_credits; the codes are in Errors and rate limits. Leave a margin of a dollar so a retry still fits.

If the balance runs short mid-project, top up before submitting the next job instead of trimming the script; the cost of a trimmed script is a message that no longer says what you meant.

Compare that with the other talking route on Sume. VEED Fabric at 720p costs $0.1875 per audio second, so $5.00 buys 26 whole seconds of it, one second fewer than standard avatar video, but Fabric uses your own recorded audio and still. Choose by who supplies the voice and the face, then by price.

• Try one 8-second standard draft: 8 x 0.184 = $1.472.
• Do not change the script after a preview unless you can afford a new preview.
• Inline captions do not create a separate billed caption job.