Custom AI influencer training is fine-tuning a model on one face so every later image and video shows the same person. Timing the spend matters more than the size of it. Follow one hypothetical page through the arithmetic: sales in, the platform fee out, and the balance that reaches the bank.
What does one month of sales look like after the fee?
A month of sales reaches your balance minus one deduction: a percentage taken from each sale. Picture a page three months old, built around one trained character. Forty members on a $12 tier bring in $480. Twenty-four pay-per-view unlocks at $7 add $168.
Gross sales for the month are $648, and training has changed none of that figure yet, only the volume of content behind it. Every number here is round and hypothetical, chosen to show the shape of the calculation.
Creator Link Studio sets its fee by plan. Creator costs $49 a month and keeps 10% from every sale, while Studio costs $199 a month and keeps 3%; on Free a creator can build and preview a page, and publishing and selling need Creator or Studio, so Free carries no fee. Card processing is billed separately by the payment network.
Applying the Creator numbers to the example produces a clean line. A 10% fee on $648 is $64.80, which leaves $583.20 in the balance. The $49 plan charge is billed on its own, as a separate monthly line, and card processing and tax are separate again.
Payouts move the balance rather than the gross figure, and the cash-out goes to your bank on request with no platform hold; ticket earnings from an event release after the event. So $583.20 is the amount actually available, not $648. Modelling training spend against the gross number is the most common way creators overestimate headroom.
Fans buy credits at ten for a dollar and spend credits on unlocks, bundles and services, while memberships are charged to a card. A $7 unlock therefore costs a fan 70 credits, and the value still lands in your balance in dollars. Pricing in dollars and letting the wallet do the conversion keeps the arithmetic above readable.
Tax takes the next slice, and training costs sit on the deductible side of the ledger. Plan charges, credits, GPU time and image sourcing are business expenses, and the IRS self-employed individuals tax center sets out the record-keeping and filing duties for anyone selling in the United States.
Offers decide whether generation volume ever reaches that first line at all. A page with no tier, no bundle and no priced post turns credits into images and stops there, the pattern described in selling without a subscription app.
When does paying for a trained character beat prompting harder?
Paying for a trained character beats harder prompting once re-rolls burn more credits than a single run, and once one face has to hold across a whole set. Below roughly twenty on-model images a month, pinned seeds and disciplined prompts usually survive, and the vocabulary for that trade-off sits in consistency terms explained.
Credits are the meter that settles the question. The Creator plan includes 500 AI credits a month, one AI influencer, 25 GB of storage, five scheduled automations and AI inbox replies. Studio raises the ceiling to 3,000 credits a month, ten AI influencers, ten team seats, 500 GB of storage and custom domains. Extra characters on Creator run $15 a month each, up to five, so testing three faces adds $30 on top of the $49 base.
Compute is the cheap input in a character run, and the research explains why. Low-rank adaptation trains a small set of new parameters instead of the full network, as described in the original LoRA paper and the PEFT conceptual guide. The costly inputs are sourcing and captioning ten to fifty varied photos, and those hours are easy to undercount.
Dataset quality decides whether you pay for one run or two. Twelve near-identical selfies produce a model that works at one angle and falls apart everywhere else, a sourcing fault that reads like a training fault. The shot mix and cull rules in how many images to use cost far less than a rerun.
Staging the spend in three steps protects you from training the wrong face. Step one is a described character and a short prompt test, using credits already included in your plan. Step two is two or three weeks of posting with one live offer, which shows whether the look holds attention. Step three is the training run, built from the strongest images step two produced.
The plan choice comes after the training decision, and it has two separate lines. The Creator plan keeps 10% of each sale and the Studio plan keeps 3%, so $2,000 of gross sales means $200 or $60 to the platform. The monthly plan price is its own line and does not change with sales.
Team seats, custom domains and the number of characters you run can move the decision before sales volume does.
Tooling choice sets the recurring line item once the first run is done. Hosted trainers, self-hosted stacks and platform features differ in what you maintain and what happens after the image exists, compared in three routes to a steady face. Keeping training next to selling is the argument for LoRA character training, which works from 10 to 50 photos or a described face and produces on-model photos and video you can price on the same page.
