Why is AI video priced per second instead of per video?

Because the compute cost scales with the number of frames produced, and a clip is just a frame count. A ten-second clip is not one unit of work that happens to be longer; it is roughly twice the work of a five-second one, and the price follows the work.

What makes AI video more expensive?

Duration first, then resolution, then the model tier. Doubling the seconds roughly doubles the cost, and raising resolution multiplies the pixels per frame across every frame. Audio, upscaling and interpolation passes are usually separate charges on top.

Should I be charged for a generation that failed?

No. A failure produced nothing, so there is nothing to pay for, and any platform worth using refunds it automatically rather than on request. It is worth checking this before buying credits anywhere, because a provider that keeps the money on failure has no incentive to fix its failure rate.

Guide

What AI video actually costs

Priced by the second, not by the clip — and the reason why tells you which settings to reach for when a project is too expensive. The more useful question is what happens to your money when a generation fails.

Why per second

A clip is a frame count wearing a different unit

Generated video is made a frame at a time. A five-second clip at twenty-four frames per second is a hundred and twenty images, produced with cross-frame attention so each one is consistent with its neighbours.

So ten seconds is not one longer unit of work. It is roughly twice the images, and because the model also attends between frames the cost tends to climb slightly faster than the duration. Pricing per second is simply the unit the work is actually done in.

This is why it is the first dial to reach for. Cutting a shot from eight seconds to four halves the bill immediately, and — as the floatiness guide argues from an entirely different direction — usually improves the shot.

What moves the number

Duration, then resolution, then the model

Duration. Close to linear, sometimes worse. The most reliable lever you have.

Resolution. Multiplies the pixels in every frame at once, so it compounds with duration rather than adding to it. Generating at a moderate resolution and upscaling afterwards is frequently cheaper than generating large, and often looks no worse.

The model tier. The spread between a fast model and a flagship one can be several times per second. A draft pass on something cheap, to check the composition and the motion, before committing to the expensive model is the single biggest saving available on a real project.

The extra passes. Audio, upscaling, frame interpolation and face restoration are usually charged separately. A quoted per-second rate often covers the base generation only, which is worth checking before planning a budget around it.

The question worth asking first

What happens when a generation fails

Generation fails more often than any marketing page will tell you. Providers run out of capacity, safety screening refuses a prompt, models return malformed output, and jobs time out. This is normal, and it is the part of the economics nobody advertises.

Which makes the refund policy the number that matters more than the rate. A platform that charges for failures has a revenue stream that improves as its reliability declines, and therefore no particular reason to fix it. One that refunds automatically pays for its own faults and has every reason to reduce them.

Here, a failed generation is refunded the moment it fails, without anyone asking. Not as goodwill — it is the only arrangement that keeps the incentives pointing the right way.

Credits do not expire. Buy them on a 30-day plan, which is cheaper per credit, or in a one-off pack - neither of them renews by itself:

$20 — 1,125 credits
$50 — 2,875 credits
$100 — 5,875 credits
$250 — 15,000 credits

What a credit buys depends on the model and the settings above, so the pricing page shows the current rates and an estimate before you commit to a generation rather than after it.

Asked often

Short answers

Why is AI video priced per second instead of per video?: Because the compute cost scales with the number of frames produced, and a clip is just a frame count. A ten-second clip is not one unit of work that happens to be longer; it is roughly twice the work of a five-second one, and the price follows the work.
What makes AI video more expensive?: Duration first, then resolution, then the model tier. Doubling the seconds roughly doubles the cost, and raising resolution multiplies the pixels per frame across every frame. Audio, upscaling and interpolation passes are usually separate charges on top.
Should I be charged for a generation that failed?: No. A failure produced nothing, so there is nothing to pay for, and any platform worth using refunds it automatically rather than on request. It is worth checking this before buying credits anywhere, because a provider that keeps the money on failure has no incentive to fix its failure rate.

Keep reading

The rest of the series

Where this comes from

Unlimited Studio is built around this one problem

Everything above is what the product does: a character is locked from eight reference angles, each generation conditions on whichever angle its prompt implies, a score checks the result against the reference, and a film is shot one frame-chained shot at a time. What it will and will not generate is set out in the content policy.

How the pipeline works What it costs