True cost per usable AI image

—

your true cost per usable image

Your situation

What are you making?

Finished images a month keepers, not attempts

How experienced are you?

Six things that quietly change what you pay

How often do you get one you'd use? most people keep fewer than they expect

How big do the images need to be? costs more every time, even the rejects

Tweaks per finished image each one is charged like a new image

Do you use it evenly each month? leftover credits usually vanish

How often does one fail or get blocked? you're normally charged anyway

Count my own time usually the biggest cost

How this is calculated

A planning model, not a quote — deliberately more honest than a per-image sticker price, but every figure is an estimate.

Attempts per keeper uses a negative binomial model: N keepers at hit rate p needs N/p generations on average, and the spread widens as p falls. Ranges shown are roughly an 80% band, so a bad month is included.
Tweaks (variations, upscales, inpainting) are added as billed generations, then everything is inflated for failed or blocked generations.
Image size multiplies credit use roughly in line with megapixels.
Plan choice is optimised per tool: every tier is tested against a 12-month usage pattern including overage, and the cheapest wins.
Expired credits are simulated month by month from your evenness setting. Unused allowance is treated as lost, which is how most plans work.
Local generation amortises hardware over three years plus electricity — cheap per image, not free.
Your time, if enabled, applies your rate to reviewing every generation, not just keepers.
Prices are approximate and change often. Commercial rights and privacy terms vary by tool and plan and are not assessed here.