How much does an AI-made short cost? A breakdown of the spend

Asking how much an AI-made short costs is like asking how much a trip costs: it depends on the destination, the means of transport and how many times you change your plans. The cost of an AI video is split into items charged by different providers, each with its own unit, and on top of that come the attempts that never get published. This guide does not give prices, because they change every month. It gives you something more useful: the list of cost items, a formula to work out your cost per published piece and the ways to put a cap on spending.

Why there is no single figure

Two shorts of the same length can cost very different amounts. Four decisions matter most:

• What you generate with AI. In cost terms, a script with a local voice and free stock clips is nothing like a video with generated scenes.
• Which model and which quality you choose. Each provider has different models and rates, and changes them often.
• How many attempts fail. A script that is redone three times, an image that is discarded or a video that is never published also have to be paid for.
• How you pay. Paying each provider per use is nothing like paying for a plan with credits, as you will see below.

That is why the figure that matters is not the price of a call, but the cost per published piece, with everything included.

The cost items, one by one

This table lists the usual cost items of an informational short. Not all of them appear in every video.

Two notes on the table. YouTube does not charge for uploading videos through its API, but it limits its use with a daily quota and, since June 2026, uploads have their own quota. X, on the other hand, charges for using its API: you buy credit in advance and each post is charged per request, and according to its pricing page a post with a link costs more than one without.

What does not appear on the provider’s invoice

Provider invoices only tell part of the story. To know the real cost, also keep track of:

• Failures and discards: generations you do not use, retries, scripts that do not pass review and videos that are never published.
• People’s time: editorial direction, review, corrections and account management.
• The machine, if you produce on your computer or your server: electricity, disk and wear.
• Fixed costs: software licences, subscriptions, domain and storage.

How to calculate your cost per published piece

The formula spreads everything spent in a period, for example a month, across what actually got published:

Cost per published piece = (Gv + Gf + H × T) / N

Gv = variable spend on ALL attempted pieces (published or not)
Gf = fixed costs for the period (licences, subscriptions, hardware)
H  = hours of human work in the period
T  = cost per hour of that work
N  = pieces published in the period

Add two indicators that will tell you where the money goes:

• Spend on failures: the variable spend on the pieces that were not published.
• Waste rate: spend on failures divided by Gv. If it goes up, the problem is not the providers’ prices but the process.

An example with hypothetical numbers

Suppose a channel uses a paid voice and generated images in some scenes, and records the following in one month:

The cost per published piece would be (120 + 60 + 10 × 25) / 30 = 430 / 30 ≈ €14.33. The spend on failures, €30, means a waste rate of 25%. Notice that, in this example, human time weighs more than any provider: that is why measuring only the price per call leads to the wrong conclusions.

Tour

Rather see it in the dashboard?

A real recording of the CLAQON dashboard, with sample data, in under a minute.

Watch the video tour

Why credits make it hard to know how much you spend

Many cloud tools do not charge in money but in credits. They are convenient, but they hide information you need in order to decide:

• A credit is not a unit of money. Each action uses a different amount, and the conversion table can change.
• The price of a credit depends on the plan. If you do not use the plan up, you pay for what you do not use. On some plans credits expire every month; on others they roll over.
• Failed attempts can use up credits. It depends on each tool, and they are not always refunded.
• You do not see the cost per item or per piece. You only see the balance go down, mixing things with very different costs.

Paying each provider directly has the opposite advantage: you see every unit and every invoice. In exchange, you manage more keys and more invoices, and you need a way to add it all up.

How to cap spending

A spend cap is a limit that stops production before it goes over. These are the ones worth having, from the most granular to the broadest:

1. Cap per piece. The most one video can spend. If a quote exceeds it, the piece is not produced.
2. Daily and monthly cap per channel. They stop a repeated error from emptying the budget overnight.
3. Reserve before spending. Calculate the cost before each expensive step and reserve the amount within the cap.
4. Alerts before the limit. An alert when you have used up a good part of the cap gives you time to decide.
5. Fail closed. If you do not know how much you have spent so far, do not spend any more until you do.
6. Do not retry blindly. If you do not know whether a paid request got through, check before repeating it.
7. Limits at the provider itself. If your provider allows it, set a spending limit in its console and turn off automatic top-up.
8. Confirmation for expensive items. Ask for explicit confirmation before generating video or before publishing on a platform that charges per post.

How CLAQON handles it

CLAQON does not use credits. You pay each provider with your own keys, at their prices, and CLAQON calculates, reserves and limits. The amounts you see in the panel are usage estimates, not invoices.

• Caps per video, per day and per month. There are budget gates before researching, writing, narrating and editing. If the budget falls short at any of them, the piece takes the “low-cost exit” before spending and the idea goes back to the queue.
• Media budget per account. Paid voice, image and video have maximums per job, per day and per month that start at zero, with no automatic top-up.
• Monthly cap per account. If you set one, it alerts you at 80%, and at 100% the account stops producing until the following month.
• Video gate. No paid video clip is requested unless seven conditions are met, starting with an approved script.
• Uncertain outcomes are not repeated automatically. A paid request without a clear response is reconciled; it is never retried on its own.
• “Spend this month” (beta). A card separates what was spent on published pieces, on failures and on pieces awaiting approval, with a breakdown by origin. If an entry is lost, the figure is marked as incomplete and the cap gate rejects new spending.
• Items with no per-use cost, if you choose them. Local voices, free stock clips, rendering on your machine, a local model with Ollama or LM Studio, and your own sources, which use no searches.
• Confirmation for anything charged separately. The paid options in Simple mode (beta) show their estimated cost and ask you to tick “I accept this cost”. Publishing to X by hand requires you to accept its cost separately.
• Prices with an expiry date. Every rate in the catalogue has a source and a date; if it expires without being renewed, no paid item is quoted.

You will find the details in spend control. And if you are wondering what changes between installing the software and using it in the cloud, read self-hosted or cloud.