Cost per finished minute of AI video at 1080p, with retakes included

The cost of a finished minute of AI video is the cost of every clip you generate, not only the clips you keep. Six 10-second clips make one minute. On Sume's Gemini Omni Flash 1.1 at 1080p, a 10-second clip bills $1.88, so a minute with no discarded takes is $11.28. If you keep one take in every 1.5 generated, you generate nine clips and the same finished minute costs $16.92. The retake rate moves the answer more than the vendor's price cut does.

Where the clip price comes from

Sume's Video Router notes list Omni 1080p at $0.15 per second from fal. Sume bills list times 1.25, which is $0.1875 per second, and rounds each clip up to the cent: 10 seconds is $1.875, billed $1.88. Runway's API pricing guide, read 2026-10-05, is a useful contrast of units: it prices models in credits at $0.01 each, so gemini_omni_flash at 10 credits per second is $0.10 per second on that page. Compare unit and resolution before you compare totals.

Other rows at 10 seconds

The same arithmetic on other models in Sume's Video Router notes gives different baselines. Gemini Omni at 720p lists at $0.10 per second, so $0.125 per second billed and $1.25 per 10-second clip, or $7.50 a minute. MiniMax H3 Max at 1080p lists at $0.16, so $0.20 billed and $2.00 per clip, or $12.00 a minute. Wan 3.0 at 1080p lists at $0.20, so $0.25 billed and $2.50 per clip, or $15.00 a minute. All of these are before retakes, and all assume you read the live pricing_skus first.

• Count generations, not seconds you ship.
• Test a draft at 720p, then render the keepers at 1080p to cut the retake cost.
• Read usage.cost from completed jobs and divide by finished minutes to get your real rate.

Keep a rolling retake rate

After a week of production, divide clips generated by clips shipped. That ratio, multiplied by the per-minute baseline, is your true cost per finished minute and the number to quote in a budget. Prices from vendors such as fal change often, and the ratio is the part you control.