What a 30-second AI ad costs on Sume: shots, voice, edit, captions

A 30-second ad made of six 5-second 720p shots, a voiceover, a timeline render and burned-in captions costs about $4.07 at the rates in Sume's code and docs as of 2026-10-04. Nearly all of it is the footage. Everything after the shots adds roughly 32 cents.

The bill, line by line

The shots use Wan 3.0 at 720p. Its billed rate is $0.125 a second, so each 5-second shot is $0.625 and six shots are $3.75. The timeline render is $0.10 per started output minute, and a 30-second cut rounds up to one minute. Captions are a flat $0.20 for a clip up to 60 seconds. The voiceover is priced per character: the pricing table lists Sonic speech at 1 credit a character with a $38 per million credits list rate, which becomes $47.50 per million characters after the 1.25 factor. A 75-word script is around 450 characters, or about $0.02.

Where the number moves

• Resolution: the same six shots at 480p are $0.0625 a second, so the footage is $1.875; at 1080p they are $0.25 a second, so $7.50.
• Retakes: each rejected shot costs its full price. Budget for one retake in four if you are new to a prompt style.
• Model: six 5-second Seedance 2.5 shots at 720p would be about $17.33, because that model is priced by video tokens at a higher rate.
• Captions: skip them and the total drops by $0.20. For an avatar video the docs say inline captions do not create a separate caption job.

A sanity check on the shot count

Six 5-second shots is a typical rhythm for a short vertical ad, but nothing in the pricing forces it. Wan 3.0 accepts clips of 2 to 30 seconds, so a 30-second ad could be a single 30-second clip at $3.75, or fifteen 2-second cuts at the same footage cost. The footage price depends on total seconds, not on how many clips you split them into. What changes with the number of clips is the editing work and the odds that one of them needs a retake.

The render and caption lines are small enough to ignore when you compare plans. The decisions that move the total are resolution, model and how many takes you throw away.

Check it before you spend

The timeline docs describe a plan call that returns estimated_cost_usd_micros without creating a job or reserving credits, so you can price the edit step for free. For the shots, the models endpoint lists prices per model. Add the two and compare against the $4.07 figure here.