A creator management agency case study: onboarding a roster onto licensed likeness, setting house rules, and adding revenue without new exposure.
Photo via Unsplash
The pitch to a creator management agency used to be simple: AI is coming for your roster's likeness, so get ahead of it. That pitch aged badly, because it never explained what "getting ahead of it" involved on a Tuesday morning with fourteen creators and a booking calendar.
This is the operational version. It walks through how an agency actually onboards a roster onto licensed likeness — the decisions, the sequence, the mistakes — using a composite case built from how the agency console gets used in practice.
Key takeaways
- Start with a restrictive house template, then loosen per creator. Never the reverse.
- Exclusivity clauses become enforced rules, which is the fastest ROI in the whole exercise.
- The bottleneck is conversations with creators, not configuration.
- Splits should be confirmed for both metered and subscription revenue before anyone signs.
- The boundary set belongs to the creator, not the agency. Write down what happens on departure.
Two pressures, arriving together.
The first is defensive. Unlicensed generation of managed creators happens whether or not anyone consents, and an agency with no documented position has nothing to point at when it does. The second is commercial: buyers who want to use a likeness legitimately increasingly need consent they can evidence, and there has been nowhere to buy it.
The demand signals are worth reading honestly. "Creator management agency" runs at roughly 140 US searches a month on August 2026 Keyword Planner data — low, spiky, and clearly a B2B term rather than a consumer one. Meanwhile "AI girlfriend generator" sits at 14,800 a month with a competition index of 1. The demand pool for AI imagery of people is enormous; the supply of licensed imagery is close to zero. That gap is the opportunity, and it is also the reason the unlicensed market is as large as it is. Agencies should also read the demand alongside the duties now attached to it — the Online Safety Act 2023 and the TAKE IT DOWN Act both shape what a buyer of your roster's likeness will be asked to evidence.
A mid-sized agency, fourteen creators, mostly lifestyle and fitness, several with active brand deals. Management is hands-on. Three creators had already found unlicensed AI imagery of themselves and asked the agency what could be done, which is what started the conversation.
Their goal was modest and sensible: establish a documented position for everyone, open a licensed route for the creators who wanted one, and not create a compliance obligation they could not staff.
They started where we suggest everyone starts — with the floor rather than the ceiling.
The house template:
Deliberately restrictive. Nobody launches wide open by accident, and every subsequent decision is a creator actively choosing to loosen something rather than failing to notice it was open.
The four non-waivable platform prohibitions — minors contexts, non-consensual framing, deceptive news contexts, criminal contexts — needed no configuration. They apply to everyone and cannot be enabled, which removed an entire category of conversation from the agenda.
This is the part that takes real time, and it is the part worth protecting.
Each creator sat down with the boundary reference open and went through their own set. The agency's rule was that nobody's profile went live until they had personally read what it said.
Six of fourteen made no changes at all. Five added contexts — fitness, fashion editorial, travel and lifestyle. Two raised the realism ceiling to photoreal, and both later dropped back to stylised after seeing sample output. One declined to license anything, which the agency treated as a completely legitimate outcome and recorded as such.
That last point matters more than it sounds. An agency that treats non-participation as a problem to solve is not gathering consent; it is gathering compliance. The configuration guide covers the field-by-field detail they worked through.
The fastest payback, and the part nobody anticipated.
Four creators had active exclusivity clauses — a sportswear brand, two supplement deals, a drinks sponsor. Those clauses lived in signed PDFs that the agency's bookings lead held in her head.
Listing excluded competitor brands in each boundary set turned those clauses into rules the system applies. A request naming a barred competitor is refused server-side, before anything is produced.
"We had been managing exclusivity by remembering it. One person remembering it, mostly. Writing it into the boundary set was the first time the obligation existed anywhere other than in her head." — composite of agency feedback, LikeWard marketplace desk
That is a governance improvement with nothing to do with AI. It just happened to arrive in the same box.
| Revenue shape | How it behaves | What to confirm |
|---|---|---|
| Metered generations | Per-image fee, split at transaction time | Does the split cover this? |
| Per-creator subscription | Recurring, better income predictor | Is the split the same rate? |
| Payout timing | Recorded per transaction in the ledger | When do funds actually clear? |
| Split changes | Configured per agency | Can it change without your agreement? |
Table: the four questions an agency should settle before onboarding a single creator. The last row is the one most often left unasked, and the only one that compounds.
One honest note on the current state: LikeWard records payout splits in the ledger, and Stripe Connect payout transfers are not yet live. Splits are computed and stored per transaction; money does not move automatically yet. We would rather say that plainly than let an agency discover it during their first reconciliation.
Three things, all instructive.
They nearly templated the ceiling. The first draft applied the house defaults as final settings across all fourteen and treated individual review as optional. Caught before launch. That version would have produced fourteen identical boundary sets and zero actual consent.
Two creators raised the realism ceiling without seeing output. Both reverted. The lesson is to show sample output at the ceiling being requested, not describe it.
Nobody tested a refusal until week three. When they did, the specificity of the message — the rule named, the values quoted — did more to build confidence than any documentation had. Test a refusal on day one.
Reasonable yes if: you have creators asking about unlicensed imagery, you manage brand deals with exclusivity terms, and you can staff the individual conversations properly.
Reasonable no if: your roster has not asked, you cannot commit to per-creator review, or you are looking for a revenue line rather than a governance position. Agencies that arrive purely for the revenue tend to template the ceiling, and that is the failure mode that turns a consent product into a paperwork exercise.
Next steps: the agency overview for how the console works, the boundary configuration guide for the fields themselves, and how to license your likeness — worth sending to creators before their conversation rather than during it.