How usage works
Your plan includes generations every month, with a number we guarantee, plus free generating on our own GPUs whenever they are idle.
The meter
Instead of a balance that ticks down, you get a meter: the share of your month you have used. Generating uses some of it, the amount depending on how heavy the model is and how busy the platform is right now. It resets on the 1st of every month (UTC), whenever your plan started.
Every plan says at least how many generations it includes. That is the worst case: the number you can run on the heaviest model your plan allows, at the busiest time of day. Almost everyone lands well above it, because a lot of generating happens while the GPUs are quiet, and on turbo and standard models that costs nothing at all.
Quiet, normal, busy
Our own GPUs run around the clock whether anyone is using them or not, so when they are idle we would rather you had them.
Quiet
Hardly anyone is generating. Turbo and standard models are free and unlimited and don't touch your meter. Heavier models, video and partner API models still draw on your month.
Normal
The usual state. A generation uses its model's standard share of your month.
Busy
Everyone is generating at once. Generations on our own GPUs use more of your month, or you can wait for things to go quiet and use none of it.
The current state is always shown on the generate page, so you never have to guess.
Heavier models cost more
A small turbo model finishes in a couple of seconds. A large one can take twenty times the GPU time for the same picture, and the biggest rent extra GPUs by the second. If both cost the same, turbo users would end up paying for everyone else, so usage scales with what the run actually costs.
| Model class | Examples | Relative cost |
|---|---|---|
| Turbo | Fast, low-step models | 1× |
| Standard | SDXL, Flux 1 dev and similar | 8× |
| Heavy | Flux 2, Qwen Image, HiDream and other large models | 45× |
| Very heavy | The largest image and editing models | 130× |
| Video and audio | Priced by the second of output | By length |
What each plan guarantees
A floor, not a ceiling. Turbo and standard generating is free whenever our GPUs are idle, on every plan including Free.
| Plan | Included generations a month, at least |
|---|---|
| Free | 180 on standard models |
| Starter | 930 on standard models |
| Hobbyist | 2,000 on standard models |
| Pro | 880 on heavy models, 5,000 on standard |
| Studio | 610 on very heavy models, 10,000 on standard |
| Max | 1,000 on very heavy models, 17,000 on standard |
| Enterprise | Custom |
Discord server plans work the same way, except the pool is shared across everyone in the server, each member can use only part of it, and it works in Discord only, never on the site.
If you run low
You'll watch your meter fill up long before it matters, and you have three ways on. None of them leave you stuck in the middle of a project:
- Wait for quiet hours. Turbo and standard generating is free while our GPUs are idle.
- Reset usage. Clears this month's usage, so your meter starts again. It costs one month of your plan, at least $5. Once per calendar month.
- Grab a pass or upgrade. A 1-day or 7-day Pro pass adds its own generations and Pro models, or you can move up a plan.
Where your money goes
Part of what you pay covers the GPUs. Another part goes to the person whose model you used: creators earn a royalty on every generation run with their model or LoRA. That's the whole point. People who publish good models should get paid when other people build on them.
Questions
Is "free when quiet" real, or is there a catch?
It's real, for the models our own GPUs run: turbo and standard. Those GPUs are paid for 24 hours a day whether they're working or not, so an idle one is money already spent. There are two catches. When everyone wants the GPUs at once they aren't quiet any more, and heavier models, video and partner API models always draw on your month, because the biggest jobs rent extra GPUs by the second.
Do unused generations roll over?
No. Your meter resets on the 1st of each month (UTC). A usage reset clears this month's usage at the moment you buy it, and it never carries into the next month.
What happens when I hit 100%?
Generating that draws on your month pauses. Turbo and standard generating doesn't while our GPUs are idle, so you can keep working for free. Reset your usage or upgrade if you'd rather not wait.
Why does the cost change during the day?
Because our costs do. When the queue is full, every generation competes for the same hardware. Charging more at the peak and nothing at the trough is what pays for a genuinely free quiet period, instead of one hard cap on everyone all the time.
Do I need to understand any of this to use Hartsy?
No. Pick a plan, generate, and watch the meter if you feel like it. This page exists because we'd rather explain how the machinery works than hide it.