TL;DR:
We've been calculating creator compensation wrong. It's meant to be 25% of what a generation costs, but it's been coming off our add-on fees too (the extra-resource fee, priority, tips), not just the base generation cost. Compensation was built before all those add-ons existed, and we never went back and updated it.
Fixing that means compensation drops for some of you. We're moving it to 25% of the base cost, the part that's actually your model. For a typical checkpoint that's around 1 buzz per generation today, and for a typical LoRA about 0.4 buzz, so it's small in absolute terms, but it's real and we're being upfront about it.
In exchange, licensing fees get a lot better. No more price cap: everyone gets the top-tier cap, so you set your own rate. The limit moves to how many new models you can put a fee on each month, by tier: Free 3, Bronze 10, Silver 25, Gold unlimited. Scale up by upgrading your tier.
Your banked buzz gets worth more too, because we stop overfilling the pool it pays out from.
And good news: you'll be able to buy memberships with buzz, rolling out later this week, so free creators can move up tiers without spending cash.
This is a proposal, not a done deal. The details and real numbers are below. Tell us where we're wrong.
The bad news first
There's a cut in here. We're correcting how creator compensation is calculated, and for some of you the per-generation payout goes down. Not burying that three sections deep.
Here's why we're pairing it with fixes instead of shipping it alone. Two things get better: the buzz you bank is worth more, and you get a way to earn that you actually control. A licensing fee you set, instead of a cut of how someone else happens to use your work.
Nothing new is getting bolted on. The way you already earn gets more accurate, more in your hands, and worth more per buzz.
How compensation works today
Most people have never seen the math, so let's start there.
Someone generates with your model. That generation gets charged some amount of buzz. Compensation takes 25% of the charge and splits it among the creators whose resources were used. Checkpoint takes the larger share, the rest splits the remainder.
The catch is what that 25% is calculated on. A charge isn't just your model. It's the base cost of the job plus add-ons: an extra fee when a lot of resources get stacked into one generation, priority, tips, licensing fees. Right now the 25% comes off all of it.
What the 25% should come out of
The only number that should drive compensation is the base cost of the generation: the cost of the actual run that used your model. On a simple image that's just the generation fee. On a big high-res ComfyUI workflow it's the cost of that whole run. Either way, that's the part that reflects your model being used, and it's exactly what compensation came out of when we first built it.
Everything stacked on top of that base is us covering our own costs, not your value. The stacking fee for piling on lots of resources is our compute cost. A tip is already paid to you in full. A licensing fee is already paid to its owner in full. Taking another 25% on top of all that means we've been handing out a slice of our own costs, and in a couple of spots paying for the same thing twice.
So compensation should be 25% of the base cost, and nothing layered on top. That's what it always meant. Back when we launched, the charge was basically just the generation fee, so 25% of the whole charge really was 25% of your model being used. Then we added the stacking fee, priority, tips, and licensing fees, and never updated the split, so it quietly kept taking a cut of things that were never yours. That's the bug. Fixing it isn't walking back the promise, it's making the math mean what it always meant.
What that looks like in practice
Take a plain SDXL generation. Base cost is 4 buzz per image, same across the SDXL family including Pony and Illustrious. Stack a few extra LoRAs in and the charge lands around 6.
Compensation is 25%, split among the creators involved.
Today: 25% of the full 6 buzz.
After: 25% of the 4 buzz base.
The piece coming out is our compute cost, not anyone's creative contribution.
This scales with the base model. A pricier base means a bigger base to take 25% of. A Krea 2 Turbo image runs around 15 buzz. A Krea 2 Raw image runs around 42, so on Raw the 25% comes off that. The more expensive the base model, the more that base-only 25% is worth to you. Which is right, because a bigger base model is more of your value, not less.
One nuance in your favor: batches get a bulk discount on the base cost, and compensation is 25% of the discounted base. Per image you earn a bit less than a one-off, but they're buying ten at once. Normal bulk economics, and it still nets out ahead of people generating one at a time.
Real numbers
Today the typical resource earns roughly half a buzz per generation it's used in. A checkpoint, which takes the larger share, earns around 1 buzz.
Flip it around and you get the useful answer: to fully replace what a typical checkpoint earns from compensation today, you'd set a licensing fee of about 1 buzz per image. For most resources, less. Uncapped Creators already using licensing fees are landing right in that range on their own.
Not waving off the concern. Just starting from the real scale instead of the worst case.
Why this is tied to what your buzz is worth
Payouts for the creator program operate as a rewards program from a shared pool, not a fixed exchange rate. When you bank buzz, your payout is your share of that pool, and your share shrinks as more total buzz gets banked. Mint more compensation than the work actually backs and everyone's buzz is worth a little less.
Creators have told us the effective value's been running low, in the 0.4 to 0.6 range. That hits hardest exactly where it matters right now: set an early-access or paid-access price with a weak buzz value and you have to charge double to land the dollars you wanted. Those are the systems we're trying to grow, and a compensation error is dragging them down.
Fair question: if everyone's compensation drops together, doesn't everyone stay in the same relative spot? For compensation alone, mostly yes. But holding the ratio isn't the goal. The goal is to stop overfilling the pool so banked buzz is worth more, then hand you better levers on top: licensing fees, paid access, early access, where you set the price and keep the value.
Compensation becomes the baseline. The fee systems become how you grow.
The other half: licensing fees you control
If you earn mostly through compensation today, the real question is whether licensing fees can cover the difference. That's the thing we want your read on. So the compensation fix comes paired with a real upgrade to fees.
Licensing fees fit what you're actually doing: selling access to your model. Compensation never did. What you earned was a percentage of how someone else used your work. More complex a workflow and the more add-on fees someone selected, the more you earned, but you had no control over that.
A fee flips that. You charge for your product at a price you set, and it doesn't matter whether it lands in a tiny workflow or a huge one.
We used compensation at the start because we didn't think people would accept licensing fees. That was three years ago. The ecosystem's settled since. Paying for access is normal now: charge if you want to, and if someone doesn't want to pay, they don't have to.
Here's the proposal:
Everyone gets the top-tier price cap. Today your max price per image is capped by membership tier, and free creators are capped low. We'd give every creator the cap the top tier has now, up to 100 buzz per image on a checkpoint. Deliberately high, high enough that in practice you price your work wherever you want. Price it too high and the market says no. That's fewer sales, not a problem we need to cap away.
Tier controls how fast you add new fees, not how much you charge. You can put a licensing fee on a set number of new models each month, and that number rises with tier. Once a fee's on a model, change the price anytime. This came straight out of creator feedback as a fairer lever than capping your price or capping how many models you can monetize total.
Membership is scale, not permission. A free creator can charge top-tier prices on their best work. Membership just lets a high-volume creator roll fees out across more new models faster.
Where we'd start the monthly limits:
Free: 3 new fee-priced models a month
Bronze: 10 a month
Silver: 25 a month
Gold: unlimited
These mirror our existing tier caps. Starting point, not a final answer. If they feel off, say so.
And if you're a free creator who can't swing a membership right now, you're not stuck. This week we're adding the ability to move up membership tiers with the buzz you earn. The path to the higher limits doesn't require your wallet. You can get there on what you make here, which is how it should work.
Why pace it at all instead of opening the floodgates? the current caps constrain the market, and that's the opposite of what we want. We want you charging what you want for your work. We also don't want the whole site to turn into a wall of paywalls overnight. Pacing how fast paid access spreads protects the experience for the people using your models and gives the market time to adjust. The monthly limit is our shot at that balance, and it's the piece we're most open to reshaping.
A few common questions
Is base creator compensation going away?
No. Compensation stays for everyone. We're only fixing what it's calculated on, moving it to the base generation cost.
Do I need a membership to earn, or to charge for my work?
No. Free creators could set fees and earn. Membership only affects how many new models you could put a fee on each month, not whether you can, and not how much you can charge.
What's the difference between a licensing fee, paid access, and early access?
A licensing fee is a small per-generation fee when someone uses your model in the generator. Paid access is a one-time price to unlock and download a model. Early access is a timed window where supporters can get in first before something goes free. They're separate tools, use whichever fits.
Is this limiting how much I can charge?
No, the opposite. We're removing the tier-based price caps, so everyone gets the top ceiling and sets their own rate. The tier now limits how many new fees you add per month, not the price.
What counts toward my monthly limit?
Each time you apply a fee to a model, a licensing fee or paid access, it counts as one against your monthly limit. Early access doesn't count toward it for now. Fee applications are spent when you use them. If you remove a fee later, you don't get that application back, so spend them on the models you actually want to monetize.
What happens to my fees if my membership lapses?
Every model you've already set a fee on stays exactly as it was. The only thing membership changes is how many new fees you could set each month. If it lapses, your monthly allowance just drops back to the free rate, nothing you've already priced changes, and nothing gets unlocked.
This is a proposal. Tell us where we're wrong.
Not a done deal. Here's our actual constraint, the mistake we think we made, and how we're thinking about the fix. We do think the current calculation is wrong, and we'd rather fix it than keep paying compensation on top of our own costs. Show us we're reading it wrong and that changes our answer.
Feedback's open through , then we'll post a follow-up with where we land.
What we'd like your take on
Now that you can see the math, does it look fair? Where would you draw the line between your value and our cost?
Monthly limit on new paid models, no cap on price: right lever, or would you shape it differently?
With fees uncapped, is there anything that stops them from replacing what compensation pays you today?
Read it. Push back. We'd rather hear it now than after we ship.