A Manga Character Now Collects a Royalty Every Time Someone Generates It

Most answers to "what happens when fans generate your character" land in one of two places. A ban, or a flat licence that trades the rights for a fee and a guess about volume.

A Japanese platform called AICU is running a third option in production. Each billable generation from a certified character reference triggers a payment to whichever rights holder the character is registered under: 40% of the gross margin left after the generation cost is deducted. Not 40% of a licence fee. 40% of the money made on that specific generation, counted per image.

How the loop is wired

The flow has four steps, and each one exists because the previous one is auditable.

1. Certify. The rights holder registers the character at cert.aicu.ai, uploading the reference image and declaring the payee. Credit and payee are separate fields in the API: original_author names the creator on the credit line, holder is who the money goes to, and they are not always the same person.
2. Generate. A fan generates the character on api.aicu.ai, spending credits from the platform's prepaid AP balance. A request with identical parameters is served from a cache: it comes back flagged as cached, is not billed, and accrues no royalty.
3. Record. Each issued licence is written to an append-only ledger, and 40% of gross margin is credited to the payee, with a monthly report. Corrections are made by appending an offsetting row rather than rewriting history.
4. Verify. Any output can be checked at cert.aicu.ai/gc: who generated it, from which reference, and how much was returned.

The character it launched with is Kurobuchi, the protagonist of YOUKAI, a school-horror manga drawn by an AI-assisted pipeline. The rights holder is Studio YOUKAI, art by Karao, original work and supervision by Dr. Shirai. According to the platform, listing periods, carry-forward of unpaid balances and monthly reporting are all part of the running system. The mechanism is covered by a granted Japanese patent (No. 7,814,797) with EU, US and Chinese applications pending.

Two details in that list are doing more work than they look like. Carry-forward means an unpaid balance is not written off at the end of the month; it rolls into the next period, which is the usual way micro-royalties evaporate everywhere else. And the verification page means the creator can audit their own statement without asking the platform to explain it. If you have ever tried to reconcile a streaming payout, you know that second one is the whole argument.

Why charge per generation instead of per licence

A flat licence has to price something unknowable. The licensor guesses how much a character will be used, the licensee argues the guess down, and the resulting number reflects negotiating power rather than usage. Small creators lose that negotiation by default, because the deal is not worth a lawyer's time.

Metering replaces the guess with a counter. A character nobody generates earns nothing. A character that gets remixed a thousand times pays a thousand times, with no renegotiation. It also removes the awkward first conversation: the price is a published percentage, so the answer to "what will this cost me" is arithmetic instead of a meeting.

The trade is that the percentage has to be small enough that a fan will still generate, which is why 40% of gross margin rather than 40% of revenue is the honest way to read it. The API documentation's own worked example splits ¥50 of margin into a 20% platform fee, the 40% royalty, and a 40% remainder that stays with the operator. Generation cost takes its bite before any of that. Shirai has said publicly that the amounts start small. Treat this as a structure that can grow with volume, not as income today.

The weak point: the meter only exists where the generation happens

Here is the part that matters for anyone planning to copy this.

A per-generation royalty is enforceable at an API, not on a downloaded file. Once a certified reference image is out, nothing in the license stops someone from taking it to a different service and generating there, where no ledger exists and no percentage is taken.

So the moat is not the licence text. The moat is the venue: the service that has the character registered, the credits users have already bought, and the workflow they are used to. GC works because AICU owns both ends, the generation API and the certification registry. An independent artist licensing a character to a third-party tool has no such control, which is the same reason image licensing has always been a distribution business more than a legal one.

Second consequence: provenance marks are receipts, not copy protection. The record logs that a generation happened and what it paid, as the issuer logged it, and the lookup resolves a licence token rather than cryptographically signing the image. AICU states that its generated output is not C2PA-compliant. That is fine, and it is worth being precise about, because "provenance" gets marketed as if it were DRM. What it actually buys is trust on the payout side. A creator who can check the issuer's own ledger no longer has to take a platform's summary on faith.

What actually changed

The framing shift is the interesting part. In the licensing model, a fan generating your character is a rights problem to be managed. In this model, that fan is a customer of your character, and the platform is the cashier.

That reframing has an uncomfortable implication. If fan generation is revenue, then a creator's incentive is to be generated, which pushes toward permissive terms and broad availability. The creators who benefit most are the ones whose characters already have an audience, and the ones who benefit least are the ones whose work is distinctive but obscure. A percentage of zero is still zero.

The second uncomfortable implication is about the reference itself. The billable unit is usage of the character, but the reference is what every generation is built from, so curating it becomes an asset decision. A sloppy reference produces worse generations, which produce fewer repeat generations, which produce less revenue. That inverts the usual advice about keeping your character art private. Under GC, the reference is the thing being licensed, and it should be built to survive being prompted by strangers.

For anyone working with original characters, four things are worth stealing regardless of whether you ever sign this licence:

Price usage, not the copyright. Selling the copyright is a one-time transaction that ends your relationship with the character. Metering keeps it.

Decide the billable unit and publish the split. "Per generation" is defensible because it is countable. "Reasonable compensation" is not a unit.

Keep a statement a stranger can check. A monthly report with a public verification path is what separates a royalty from a promise.

Assume you need distribution to make any of it real. A metered royalty on a platform with no traffic is a nice document.

The part to be skeptical about

There is no published volume data here, and the launch character belongs to the same organisation that built the platform. That makes GC a credible proof of mechanism and not yet proof of a market. The genuine open question is whether fans will pay a metered premium for a specific character when a generic one is free, because that is the only thing that makes the percentage worth anything. Character identity is the product being tested, not the payment rail.

What is no longer arguable is that the rail exists. A year ago, the serious position on AI derivative work was that licensing was theoretically possible and practically unadministrable, because nobody could count the uses. Someone built the counter, attached a percentage to it, and started paying artists from it in production. Whatever you think the number should be, the argument now has a denominator.