How to stop free credit abuse in an AI headshot generator
A free headshot pack pays for a training run on every new account. What AI headshot generator credit abuse costs, and which basic fixes break.
$4,320 is what you paid last month to train models for people who were never going to buy from you.
That's a scenario, so swap in your own numbers once you've seen how it adds up. You run an AI headshot generator. Someone uploads 15 selfies, you train a small model on that one face, and it turns out headshots good enough for LinkedIn. Every new account gets 12 free credits, one per headshot. Here's what those 12 credits cost you, using assumptions you should replace:
Training run: 20 GPU minutes x $0.09 = $1.80
Generation: 12 headshots x $0.04 = $0.48
Upscaling: 12 headshots x $0.01 = $0.12
One free pack = $2.40That $1.80 is what makes AI headshot generator credit abuse so expensive. A model trained on one face is no use for the next face, so every new account buys a brand new training run, and you've paid for it before anyone has seen a single photo.
Now say four people have noticed. Each one sells "AI headshots, $5" on a freelance marketplace. They take 15 orders a day, open a new account on your site for each order, upload the client's selfies and hand over your free pack.
4 sellers x 15 orders a day = 60 free packs a day
60 packs x 30 days = 1,800 free packs a month
1,800 packs x $2.40 = $4,320 a month, paid by you
1,800 orders x $5 = $9,000 a month, paid to themThat's $144 leaving your account every day (60 × $2.40). Your GPU provider charges you as the runs finish. The sale that's supposed to pay for a free pack comes later, if it comes at all, and these never will. The headshots have been downloaded, the GPU time is used and the account's been left behind, so there's nothing to take back.
Here's the rest of your month. 1,500 honest free packs turned into 210 sales of a $29 pack of 100 headshots, and each of those costs you $6.80 to make ($1.80 + 100 × $0.05). This is before payment fees.
Profit per sale: $29 - $6.80 = $22.20
Profit from sales: 210 x $22.20 = $4,662
Honest free packs: 1,500 x $2.40 = $3,600
Left over: $4,662 - $3,600 = $1,062
After the four sellers: $1,062 - $4,320 = -$3,258Just covering the sellers takes 195 of your 210 sales ($4,320 ÷ $22.20, rounded up). A product that was adding $1,062 a month is now losing $3,258, and if you've got $40,000 in the bank, that's a little over 12 months of runway ($40,000 ÷ $3,258). 210 sales from 3,300 free packs reads as 6.4%, when your honest conversion is 14%, and a number like that can send you off to spend two weeks retuning a model that was fine.
Paying customers feel it in the queue. Two GPU workers train 6 models an hour (2 × 3 runs). Say the sellers work through their orders in the evening, and all 60 packs land between 10 p.m. and midnight. That's 10 hours of training. A customer pays you $29 at 11 p.m. for a job application that closes at midnight, and your page says about 25 minutes. By then 30 free packs have arrived and 6 are done, so 24 are ahead of them: 24 ÷ 6 = 4 hours. Their headshots turn up around 3:20 a.m. You refund the $29 and you've still spent the $6.80. They only needed headshots once, so there's no next time to win them back.
That leaves you with a bad choice. Keep paying $4,320 a month, or switch off the free pack that produced all 210 of your sales.
Why free credits get farmed in a headshot app
Most people need a headshot about once, so 12 free credits can be everything they wanted from you. Someone who didn't like their first 12 opens a second account and tries again. A seller treats your free pack as stock. Neither sets anything off, because each account trains once, on real photos, and stays inside its limit. It's credit farming at an unusually high price per account.
The obvious fix is a smaller free pack, and this is where the training run bites. Cut the pack from 12 headshots to 4 and you've taken away two thirds of what people get, but only $0.40 of what you pay, which is one sixth.
The bars are the whole pack, and the line is the training run you pay for at any size. If a seller's clients still expect 12 headshots, a 4-headshot pack means 3 accounts, 3 training runs and $6.00 (3 × $2.00) for what used to cost you $2.40.
You've also got people's faces on your servers, so it's tempting to use them here too. They were uploaded for one job, making headshots, so anything else is a question for your lawyer first.
Stop headshot credits leaking before training starts
Grant credits at the last moment. Nothing expensive happens at signup. Grant the free pack once the email's confirmed, the photos are in and training is about to be queued, so an account that stalls costs you one email.
Allow one free training run per account, in the database, so a double click or a retried job can't start a second $1.80 run.
Give paying customers their own GPU worker. That 11 p.m. customer would have had the 25 minutes you promised.
Set a daily budget for free training. At 80 runs a day, the most you can lose is $192 (80 × $2.40). Make it a hard cap, and tell anyone past it they're in tomorrow's queue.
Watermark the free pack. A $5 client is unlikely to accept one, so the pack is worth less to a seller. You've still paid the $1.80.
Those limit the damage. They don't reduce how many packs go to repeat accounts.
The basic checks, and how each one breaks
| Basic fix | What it catches | What it misses | Who it wrongly blocks | Cost to keep working |
|---|---|---|---|---|
| Email verification and unique emails | Typos, dead addresses, the same address twice | A new real inbox per order | People whose code lands in spam | Deliverability work |
| Disposable-domain blocks | Known throwaway services | New domains and ordinary mailboxes | People using privacy forwarding | A list someone updates forever |
| Gmail dots and plus addressing | One inbox written several ways | A second real mailbox | A family sharing one inbox | Upkeep as mail services change |
| CAPTCHA | Simple scripts | A seller uploading by hand | People using screen readers | Fees and lost signups |
| Phone or card requirement | Casual second accounts | Anyone with another number or card | People who want to see 12 photos first | Text fees and fewer sales |
| One per IP and VPN blocks | Repeats on one connection, known VPN ranges | Any other network | A company trying you for its whole team | Exceptions, tickets and a list to maintain |
| Rate limits | Bursts from one account | 15 orders a day on 15 accounts | Real users in a busy hour | Tuning per endpoint |
| Usage caps | Spend past your limit | Who gets the packs under it | Honest users arriving after the cap fills | Deciding when free stops each day |
| Smaller packs | $0.40 a pack | The $1.80 training run | Every honest user | Fewer people reach paying |
| Manual review | What a person can spot | Anything after hours | Whoever looks odd to a tired reviewer | 110 minutes a day of strangers' selfies |
Sellers get past almost every row. Each one is a real person on a real connection, uploading real photos by hand, 15 times a day, so a CAPTCHA has nothing to catch. Email has its quirks as well: Gmail ignores dots, and plus-tagged mail reaches the same inbox, so one mailbox can look like many addresses.
The rows that do bite often bite the wrong people. An office manager testing you for a team of 30 arrives on one connection, quite possibly a company VPN, and that's $870 of sales (30 × $29) your rule just turned away.
Stack them all and your signup asks for a confirmed email, a puzzle, a phone number and a card, just so someone can see 12 photos of themselves. It would slow the sellers down. It would also cut the free packs behind all 210 sales. Every row asks about a signup: is the address real, is the visitor human. What costs you $2.40 comes later, with the photos in and the GPU about to start.
Evaluate the claim before the GPU starts
Fresh accounts are the sellers' whole business model, and with AbuseZero, creating a fresh account isn't enough to erase previous activity. It recognizes repeat devices across new accounts, even when email addresses keep changing. A one-per-IP rule stops at the address. AbuseZero sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. Our multi-layer AI risk analysis examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. We've also got a proprietary algorithm, with AI that checks the claim again. No row in that table comes close.
Your server makes one API call once the selfies are in, and gets a clear allow or deny under rules you control. A free pack can be once per user, ever. Observe mode shows you what would be denied before anyone is blocked, so you can see how many free packs look like repeat claims first. The quickstart protects your first benefit. You don't have to build your own abuse prevention system, and you can get back to making better headshots. Give every deny a paid pack and a way to reach you, because some of the people you turn away will be real.
Frequently asked questions
How much does AI headshot generator credit abuse cost per account?
Add your training run to your cost per headshot, multiplied by the free credits. In this scenario it's $2.40, and $1.80 of that is the run. Work out your own.
Should an AI headshot generator have a free pack at all?
Keep it if it sells your paid pack. In this scenario, honest free packs left $1,062 a month after their own cost. If yours don't, read should you remove your free tier.
Is making a second account to retry always abuse?
No. Someone unhappy with their first 12 may just need better selfies. Offer a few free re-rolls on the same account: if you've kept their trained model, each one costs you $0.05 and no new $1.80 run.
Spend your GPU budget on future customers
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.