Retake budget for a 500-prompt re-render: 15, 25 or 35 percent extra

Budget 500 prompts as 575, 625 or 675 renders. At 8 seconds and 720p that is $375.00 to $1,512.50 on Sume with one retake in four. Table by model.

5 min readSume
All posts

A 500-prompt re-render at 8 seconds and 720p costs $300.00 to $1,210.00 on Sume with no retakes. With one retake in four (625 renders) it costs $375.00 to $1,512.50, with wan-3.0 at $625.00. Plan on a retake rate, because an old prompt written for sora-2 will not always produce a usable clip on a new model the first time.

OpenAI removed the Videos API and the sora-2 family on 2026-09-24 and listed no replacement. A prompt tuned to one model's habits is the likeliest to need a second try on another, so the retake line belongs in the budget from the start.

Three retake rates

The rates below are planning examples: 15 percent adds 75 renders, 25 percent adds 125 and 35 percent adds 175 to a base of 500. Each retake is a full new generation at the same price.

500 clips at 8 s, 720p (768p for minimax-h3) plus retakes, billable on Sume (read 2026-10-08)
ModelNo retakes+15% (575)+25% (625)+35% (675)
minimax-h3 (768p)$300.00$345.00$375.00$405.00
wan-3.0$500.00$575.00$625.00$675.00
gemini-omni-flash-1.1$500.00$575.00$625.00$675.00
kling-3, audio off$560.00$644.00$700.00$756.00
seedance-2-mini$760.00$874.00$950.00$1,026.00
seedance-2-fast$1,210.00$1,391.50$1,512.50$1,633.50

What is and is not a retake

A retake is a clip that completed and that you rejected, so you pay for it and pay again for the replacement. A job that fails is different: the Sume docs say failed jobs and failed queue admission release or refund the reservation, so a failure does not count against the retake line.

The arithmetic for one row: wan-3.0 at 25 percent is 500 x 1.25 = 625 renders, 625 x 100 cents = 62,500 cents, or $625.00. For seedance-2-fast the same 625 renders at 242 cents come to $1,512.50. The retakes alone, 125 renders, cost $125.00 on wan-3.0 and $302.50 on seedance-2-fast.

Keep retakes safe and visible

Each retake needs a new idempotency key. If you reuse the key of the first take with the same payload, Sume returns the original job, and you get the clip you rejected instead of a new one. Add a take number to the fields you hash into the key, and keep it in your table next to the job id and usage.cost.

Track the retake rate per model as you go. If the first 50 clips on a model show 40 percent retakes, the table above is optimistic for that model; stop and look at the prompts before you spend the rest. The no-seed rule also means that a retake is a new sample, not a repair of the old clip.

  • Log take number, model, usage.cost and the reason for the reject.
  • Cap the retakes per prompt at two, and escalate to a human rewrite after that.
  • A change of model for a hard prompt is a new row, with the new model's price.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume