How to limit free AI image generations per person
Free image generation abuse is paid for before you see it. How to price a free image, cap the spend and why a limit per account isn't per person.
The invoice open on your phone has a total on it that can't be right. Your model provider sent it this morning, the 1st, and you opened it in bed expecting something close to last month's $3,240. The total is $12,840.00, and your card gets charged on the 3rd.
Treat this as a worked example and put in your real figures. Every account on your image tool gets 50 free generations a month, and each image costs you $0.04 at your model and resolution. You open your database and find 4,800 free accounts that used all 50.
4,800 accounts x 50 images = 240,000 images
240,000 images x $0.04 = $9,600
$3,240 normal bill + $9,600 = $12,840Not one of those accounts went over its limit, and not one has paid you. That's free image generation abuse, and what makes it so hard to live with is the order things happen in. The GPU ran, the image was delivered, and now it's sitting on someone else's drive. You can delete all 4,800 accounts before breakfast and you'll still owe the $9,600.
Your revenue doesn't show up in time to help. 190 subscribers at $29 bring in $5,510 a month, so this one invoice is 2.3 months of everything they pay you. After a normal $3,240 bill you keep $2,270, which means the farmed images ate more than four months of that ($9,600 ÷ $2,270 = 4.2). If your company burns $6,000 a month, you've just handed 1.6 months of runway to people who wanted free pictures.
Then you open the support inbox. Your provider's per-minute limit covers your whole account, so on the 27th, when those free accounts were busiest, your paying customers' jobs started timing out. Eleven of them wrote in. One had a client presentation that afternoon and asked why they were paying $29 to wait behind free users. You refund all eleven, which is another $319 (11 × $29). Three cancel anyway, and $1,044 of next year's revenue goes with them (3 × $29 × 12).
The cost climbs in a straight line with every account, at $2.00 for each one fully used (50 × $0.04):
It's 6:10 a.m. by now, and there's one more thing. Your free allowance reset at midnight. All 4,800 accounts have 50 new images, and your usage page shows 18,400 generated since then, which is $736 (18,400 × $0.04) spent while you were asleep. You can shut off the free tier, the one that brought you nearly all 190 subscribers, or you can leave it open and start paying next month's invoice this morning.
Why free image generation abuse gets past a limit per account
Your limit did what it was told. It counted images per account, and all 4,800 accounts stayed at 50. What you meant was 50 images per person, and nothing in your database records a person. It records rows, and a new row costs someone a minute and an inbox.
Image tools get hit harder than most because the output is the entire reward. Nobody has to come back to keep a product photo or a thumbnail, so there's no reason to stay in one account. One person with 40 accounts gets 2,000 free images a month (40 × 50). That costs you $80 (2,000 × $0.04), or nearly three months of your $29 plan. That's credit farming, with a result that's easy to download and reuse.
The obvious fix is a smaller number. Cut 50 to 10 and each account costs $0.40 instead of $2.00. But image prompts take a few tries, so an honest newcomer can run out before seeing anything worth paying for, while the person with 40 accounts opens 200 and collects the same 2,000 images (2,000 ÷ 10). Set the allowance from what a first good result takes, because shrinking it mostly moves the problem onto your honest users.
How to limit free AI image generations without a surprise bill
These steps won't tell you who's who. They decide what a free image costs you and how much a bad month can take.
Price one generation honestly. Count retries, variations and upscales. A button that returns a grid of four costs $0.16 a click at $0.04 an image, so give free users one image per click. Then work out what a free user costs you.
Count on the server, before the model runs. A counter in the browser can be cleared by the person it's counting. Take one image from the allowance in a single statement, and call the model only if a row comes back:
update free_image_allowance
set used = used + 1
where account_id = :account_id
and billing_month = :billing_month
and used < 50
returning used;Two requests landing together can't both take the last image. Give an image back only when the generation failed on your side.
Make free images cheaper to produce. Use a cheaper model, a lower resolution and no upscaling on the free plan. At $0.01 an image, the same 240,000 images cost $2,400 instead of $9,600, which is $7,200 less. Keep it good enough to sell the paid plan.
Watermark free output, and know what that buys you. A visible mark makes an image less useful for clean commercial work, and it advertises your tool every time someone shares one. It does nothing for the bill, because you've paid for the generation before the mark goes on. An honest evaluator can't put a watermarked mockup in front of a client.
Give free traffic its own budget and its own queue. A $100 daily limit on free generations caps the month at $3,000 ($100 × 30), and a separate queue keeps paying customers' jobs out of the pileup. The catch is that farmed accounts can spend the $100 by 9 a.m., and every honest newcomer for the rest of the day gets "come back tomorrow". Here's how to set a hard spending cap.
You've capped what a bad month costs, but you still don't have a limit per person. The next thing to try is counting something closer to one.
The basic checks that try to count people, and how each one breaks
Each of these catches someone. Turn on email verification and unique emails at the very least, and run no generation until the email is confirmed, so an account that never confirms costs you one email instead of $2.00. Read the last three columns before you trust the rest.
| Basic check | Stops | Misses | Wrongly blocks | Upkeep |
|---|---|---|---|---|
| Email verification | Typos and dead inboxes | Any second real inbox | People whose code went to spam | Deliverability work |
| Unique emails | One address used twice | Every other address | Almost nobody | Very little |
| Disposable-domain blocks | Throwaway services on your list | New domains and ordinary mailboxes | People using privacy forwarding | A list that goes stale |
| Gmail dots and plus tags | One Gmail inbox written several ways | A second real mailbox | People who sort their mail with plus addresses | Upkeep as mail services change |
| CAPTCHA | Simple scripts | People making accounts by hand | Screen reader users | Fees and lost signups |
| Phone requirement | Casual second accounts | Anyone with another number | People without a reliable mobile | A fee for every text |
| Card requirement | Casual second accounts | Anyone with another card | Anyone who wanted to try one image first | Far fewer signups |
| One per IP address | Repeat signups on one connection | Any other network | Offices, campuses and households | Exceptions and tickets |
| VPN blocks | Known VPN ranges | Everything off your list | Staff on company VPNs | Another list |
| Rate limits | Bursts from one account | 40 accounts going slowly | Real users with a deadline | Tuning per route |
| Usage caps | One account overrunning | Each new account's fresh cap | Anyone who hits it mid-project | Picking the cutoff |
| Smaller allowance | Some cost per account | More accounts make up the difference | Every honest newcomer | Fewer good first images |
| Manual review | What a person can spot | Anything after hours | Whoever looks odd to a tired reviewer | Hours every month |
Email shows the problem most clearly. Gmail treats dotted versions of an address as the same address, and mail sent to name+anything@gmail.com still reaches name@gmail.com, so one inbox can look like many addresses. A unique-email rule compares text, so it sees different people.
The upkeep is what wears you down. A VPN block catches a customer's company network, and you find out from a ticket. A CAPTCHA costs you some of the people who came to make one picture of their dog, and those people were the point of the free tier. Manual review at one minute an account is 80 hours for 4,800 accounts (4,800 ÷ 60), and whoever does it is looking at accounts whose 240,000 images already exist.
Stack them all and someone trying your tool for the first time confirms an inbox, solves a puzzle, waits for a text and types in a card number before seeing a single image. The person after 2,000 free images will put up with that. The curious newcomer won't.
So write down what one person is entitled to, keep the budget from step five and watch your free image spend every month. Expect some of it to leak, because that's about as far as checks you build yourself can go.
When the free allowance gets claimed again
Every row in that table knows one thing about a signup. AbuseZero starts where they run out. Its multi-layer AI risk analysis examines 140,000+ signals, detects linked claims and flags abuse that single checks miss, and those signals are read together, never one at a time. You also get a proprietary algorithm, with AI that checks the claim again. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing, and creating a fresh account isn't enough to erase previous activity. The lists stop being your job as well, because it checks mailbox aliases, disposable domains and mail routing, and sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. You don't have to build your own abuse prevention system.
Your side is one API call before the free images go out, and you get a clear allow or deny under rules you control. Set up the monthly allowance as a benefit with eligibility of once per calendar month, then use Observe mode to see what would be denied before anyone is blocked. Your counter, budget and queue stay where they are. The quickstart walks you through your first benefit.
Frequently asked questions
Do watermarks stop free image generation abuse?
No. They lower what a free image is worth to some people, but the generation has already run and you've already paid for it.
Should free images refresh daily or come once at signup?
A single grant limits what an account can ever cost. A refresh multiplies what every extra account collects. Here's how daily and signup credits compare.
Can I get my money back after a wave of farmed generations?
Don't count on it, because the generations ran. You can remove unused allowances so the loss stops growing. See how to take back free credits after an abuse wave.
Protect your next free allowance
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.