What One AI Image, Video or Song Actually Costs

Almost nobody publishes what a single AI generation costs. Tools quote a monthly subscription and a vague allowance — "500 credits", "unlimited standard generations", "fair use" — and the price of the thing you actually wanted stays hidden behind a plan name.

It does not have to. Some platforms publish a per-model credit table and sell credits at a fixed rate, which means the cost of one image, one clip or one song is a multiplication you can do yourself. This piece does that arithmetic on a platform that exposes the numbers, then shows you how to run the same check anywhere else.

Why a subscription hides the number you need

A monthly plan bundles three separate things: access, an allowance, and a per-unit rate. Vendors quote the first, imply the second and almost never state the third. That is convenient for them and expensive for you, because the two questions that decide whether a tool is worth paying for are:

• What does one finished output cost me?
• How many will I actually make this month?

Multiply those and you have your real bill. A $29/month plan is excellent value at forty images and terrible value at three. Without a per-unit number you cannot tell which you are, and the plan page is designed so you cannot work it out.

Our guide to how AI agents are priced covers the four pricing models and when each one suits you. This piece is narrower and more practical: it is about getting to an actual number in currency.

A worked example, using published rates

Alexiuz runs a set of generation tools on one shared credit balance, and — unusually — its per-model prices are readable rather than marketing copy. Two published facts make the arithmetic possible.

First, the credit rate. Their credits page states: "Min $10 · Max $1,000 · Pay $10, get 1,000 credits", with a first-purchase bonus that turns the opening $10 into 1,100. So 100 credits = $1, and one credit is one cent. There is no monthly plan to fold in — their pricing URL returns a 404, because credits are the whole model.

Second, the per-model tables. Their Generor generator page carries its own price lists in the page source, one per output type. Read on 23 August 2026:

The medians are the useful column. A typical image is 9 credits, or nine cents. A typical video clip is 14 credits, or fourteen cents. A typical song is 40 credits, or forty cents. Those are prices per generation, not per month.

Speech is metered differently — by characters rather than per run — at between 10 and 100 characters per credit depending on the voice. At the cheap end that is roughly 10,000 characters per dollar; at the expensive end, about 1,000. The voice list names eight underlying providers, including ElevenLabs, Deepgram, Google, xAI and Hume, so what you are choosing between is largely whose engine you want at what markup.

What that arithmetic actually tells you

Run your own volume through it and the answer usually surprises you in one direction or the other.

• Forty images a month at the median rate is 360 credits, or $3.60. Against a $29 subscription, the credit model wins by a distance.
• Three hundred images a month is 2,700 credits, or $27 — and now the subscription is competitive, because flat plans are built for heavy users and credit models are built for light ones.
• A minute of video assembled from several short clips at 14 credits each lands in the low tens of cents, but the expensive models cost six times the median. On video, the model you pick matters more than the platform you pick.

That last point is the one people miss. Within a single tool the spread between the cheapest and dearest model is 9× on images, 30× on video and 8× on music. Choosing a mid-table model instead of the flagship is usually a bigger saving than switching platforms, and on most work nobody can tell the difference in the output.

The stack problem

The other reason per-unit pricing matters is that creative work rarely needs one tool. A short promotional video might need an image, a voiceover, a music bed and an edit — four products, and on the subscription model, potentially four monthly bills for work that takes an afternoon.

A shared balance changes the shape of that. On Alexiuz the same credits spend across the whole set, which currently runs to twelve services:

The honest framing is not "this replaces your stack". It is that paying per output removes the decision about whether a tool is worth keeping, because an unused tool costs nothing. That is the actual advantage of usage pricing, and it is worth more to occasional users than any feature list.

Where usage pricing bites back

It cuts the other way too, and this section is the one to read twice.

• There is no ceiling. A subscription caps your monthly downside; credits do not. If you automate a loop that generates in bulk, the spend goes with it. Decide a monthly number before you start and top up to it deliberately rather than whenever you run dry.
• Failed and discarded generations still cost. Per-output pricing charges for the output, not for the one you kept. Iterating a prompt eight times to get one usable image costs eight times the table price, so the real per-keeper cost is your credit rate multiplied by your hit rate. Budget on keepers, not attempts.
• Prepaid balances are not refunds. Buying $1,000 of credits at a bulk moment is only cheaper if you spend them. Money sitting in a balance you never use is worse value than any subscription.
• Published rates move. The tables above were read on one day from one page. Re-read them before making a decision that depends on the number.

How to run this check on any tool

The method generalises, and it takes about five minutes.

• Find the currency rate. How many credits, tokens or units does one unit of real money buy? If a vendor will not tell you, that is itself information.
• Find the per-model table. Often it is in the generator UI rather than the pricing page — a model picker that shows a cost next to each option, or a price list in the page source.
• Multiply by your real volume, not your aspirational one. Count what you made last month, not what you intend to make.
• Compare against the subscription at the same volume, and check both ends: what you pay on a quiet month and on a busy one.

If a tool cannot survive that arithmetic, no feature list rescues it. Our honest guide to AI tools makes the same argument about capability claims — and free AI agents 2026 is worth reading first if your volume turns out to be low enough that you should not be paying at all.

The Bottom Line

Ask for the per-output number and refuse to decide without it. On the platform costed above it works out at roughly nine cents an image, fourteen cents a video clip and forty cents a song at median model rates, with a 9× to 30× spread between the cheapest and dearest model in the same category — which means model choice is usually a larger lever than platform choice.

Usage pricing suits light and irregular work, because nothing is charged when nothing is made. Subscriptions suit steady heavy work, because the per-unit rate collapses at volume. The crossover is a number you can calculate in five minutes, and almost nobody does.

Prices and model tables were read on 23 August 2026 and change without notice. Verify current rates on the vendor's own page before committing money.

FAQ

How much does one AI-generated image actually cost?

On the platform costed here, the median model charges 9 credits and credits are sold at 100 for $1 — so about nine cents per generation, with the range running from 1 credit to 27 credits across 38 models. That is per image produced, not per month. The figure most people should care about is higher than the table price, though, because you pay for discarded attempts too: if you need four tries to get a keeper, your real cost is four times the listed rate.

Is credit-based pricing cheaper than an AI subscription?

It depends entirely on volume, and the crossover is calculable. Credits win on light and irregular use because an unused month costs nothing; subscriptions win on heavy steady use because the effective per-unit rate falls as you use more. Work out your monthly output count, multiply by the per-generation rate, and compare that against the subscription price. If you are near the crossover, prefer credits — they carry no commitment.

What is the catch with usage-based AI pricing?

There is no spending ceiling. A subscription caps your monthly downside; a credit balance does not, and automated or bulk generation can drain one quickly. Failed generations also cost the same as successful ones. Set a deliberate monthly top-up figure, treat it as the cap, and budget on the outputs you keep rather than the ones you attempt.

Does the model I choose change the price much?

More than most people expect. Within one platform the spread from cheapest to dearest is roughly 9× on images, 30× on video and 8× on music. Picking a mid-table model rather than the flagship is usually a bigger saving than moving to a different vendor, and on routine work the quality difference is often invisible. Check the per-model cost in the generator's model picker before you run a batch.