AI Image Generation Cost Calculator

Pick a model, plug in how many images you plan to generate each month, and see the per-image cost, monthly bill and annual total. Then look at the comparison table and see what the same workload costs everywhere else. Useful when you are about to commit to a generative-image workflow and want to know what the credit-card statement will say.

Explain like I'm 5 (what even is this calculator?)

Every AI image you generate costs the provider a tiny slice of GPU time, and they pass that cost on to you. Some, like DALL-E and Imagen, charge per image. Some, like Midjourney, sell you a monthly bundle. Some, like Replicate-hosted Stable Diffusion XL, are basically rounding error per image. This tool turns "I want to make 2,000 product photos a month" into a real money figure on each provider, so you can pick the cheapest one that produces the quality you actually need.

Calculate

Pick a model, set a monthly volume, then press Calculate.

Prove it

Cost is per-image price multiplied by your monthly volume; annual is monthly times twelve. Pricing is the standard list price published by each provider on the Last verified date. Midjourney is sold by subscription, so the per-image figure is the Standard plan ($30/month, ~3500 fast images) divided down. Volume discounts, free credits, upscales, variations and currency conversion fees are not modelled.

Useful? Save this calculator: press Ctrl + D to bookmark it.

What this calculator is actually doing

An image-generation API bills you per image, give or take. The maths is short: per-image price multiplied by volume gives you the monthly bill, and that times twelve gives you the annual. The hard bit is finding an honest per-image figure for each provider, because they price in different shapes (per call, per credit, per subscription) and they all change their list prices a few times a year.

The figures here are baked into the page on the Last verified date you can see in the result block. No live API calls, no surprise rate-limit failures, no way for a popular calculator to spin up a six-figure invoice on someone else's bill. If a provider drops their price tomorrow, this page is wrong tomorrow afternoon, and you should sanity-check against the provider's own pricing page before signing off real spend.

Honest caveats, in plain English

A few things this calculator does not model, all of which can move the real bill in either direction.

Subscription versus pay-as-you-go

DALL-E 3, Imagen 3 and the Replicate-hosted models are pay-per-image. Midjourney is a flat monthly subscription with a fast-image quota. The calculator treats Midjourney's effective per-image cost on the Standard plan as $0.0086, which is honest if you stay inside the fast quota and a bit optimistic if you blow past it. If you only generate 50 images a month, your effective Midjourney cost is $30 ÷ 50 = $0.60 each, not $0.0086. That is the trade-off of subscriptions.

Image size and quality

Prices here are for 1024x1024 standard outputs (or each provider's closest equivalent). Larger sizes, HD modes, multiple variations per call, and upscales all cost extra on most platforms. If you are going to generate 1792x1024 hero images at HD on DALL-E, expect to pay roughly three times the figure here.

Failed and rerolled outputs

This is the one nobody wants to talk about. Cheap models often need two or three attempts to produce one usable image. SDXL at $0.0023 is brilliant on paper, but if your real ratio is one keeper per three rerolls, your effective cost is closer to $0.007. Track your keep rate before you commit to a model on price alone.

Free credits and tiers

Replicate gives new accounts a small free monthly credit. Vertex AI offers a free trial budget. OpenAI hands out occasional promotional credit. None of that is modelled here, because once you are in production those one-off credits are noise.

How to read the comparison table

The table sorts cheapest first. Useful as a starting point, not as the final answer. The cheap rows are cheap because the model is open-weights and runs on commodity hardware. The expensive rows are expensive because the model is doing something the cheap ones can't yet, or because someone is paying for a brand. Run a representative test on each candidate before committing your spend to the cheapest row.

If two providers are within a few percent of each other on cost, your tiebreaker is usually quality, latency, or which one your team already has accounts and observability set up for. Switching cost is real, and almost never worth it for a tiny price delta.

Trimming an image generation bill, in rough order of impact

If the number this calculator produced gave you a fright, the highest-leverage moves are usually:

  • Generate fewer. Most teams over-generate. A clear brief and a tighter prompt cuts the reroll rate, which cuts the bill in proportion.
  • Drop down a tier on the easy work. SDXL or Flux Schnell handle hero photography stand-ins, blog illustrations and social-media filler perfectly well. Save the flagship models for the things that genuinely need them.
  • Cache the assets you generate. If five different blog posts need the same generic "person at a laptop" hero image, generate it once and reuse it. The cheapest image is the one you don't generate twice.
  • Self-host SDXL if your volume justifies it. At a few thousand images a month, hosted SDXL on Replicate is the right answer. At a hundred thousand, a single H100 on RunPod or a dedicated GPU on a hyperscaler will be cheaper, with the trade-off that you now own the operations.
  • Use real photography for anything important. A good stock photo is still a few quid. AI is better when the alternative is "no image at all", not when the alternative is "a properly shot product photo".

Related calculators

Image spend is one workload. These cover the rest of the AI cost picture.

Frequently asked questions

Why does Midjourney show a per-image price when it is sold by subscription?

Because the comparison only works if every model is priced on the same axis. The Midjourney Standard plan is $30 a month for roughly 3,500 fast images, which works out at about $0.0086 per fast image. If you generate fewer than that, your effective per-image cost is higher; if you stay inside the fast-image quota, the figure here is honest. Heavy users will spill into relax-mode or buy more fast hours, which the calculator does not model.

Are these the prices I will actually pay?

They are the public list prices on the Last verified date shown on the page. They do not include volume discounts, free credits new accounts get on Replicate or Vertex AI, currency conversion fees if you are billed in a non-USD card, or upscales and variations on Midjourney. Treat the number as a sensible estimate, not an invoice.

What size of image are these prices for?

The standard 1024x1024 image, or each provider's nearest equivalent. Larger or more detailed outputs cost more on most platforms. If you are generating big landscape images at scale, push the per-image price up by 50 to 100 percent before signing off the budget.

Why is Stable Diffusion XL so much cheaper?

Two reasons. First, SDXL is open weights, so providers like Replicate compete on margin alone rather than carrying a research bill. Second, SDXL runs in seconds on cheap GPUs. The trade-off is output consistency: SDXL needs a careful prompt, the right sampler and often a few rerolls before it produces what you wanted. Cheap per image is not the same as cheap per usable image.

How often is the pricing on this page updated?

The Last verified date on the page is when the prices were last checked against each provider's official pricing page. The figures are baked in so the page loads fast and works without any API calls, which means they can drift if a provider changes their list price. Sanity-check against the provider's own pricing page before signing off a meaningful spend.

Does this calculator send my prompts anywhere?

No. Everything runs in your browser. Nothing leaves your device, because there is no server to leave to.

Last updated 29 April 2026 by Andrew Laws.