Why We Run Our Own Models Instead of Reselling an API

Most background removal products aren't really products — they're a thin interface wrapped around someone else's paid API. That's a reasonable way to ship fast, but it means every image you process has a marginal cost, and that cost gets passed straight to you, usually with a markup.

The problem with metered pricing

Metered pricing sounds fine at low volume and turns into a real problem the moment a team actually depends on the tool. A catalog refresh, a busy onboarding month, a batch job that runs nightly instead of on demand — all of that scales your bill in lockstep with your usage, right when the tool is proving most useful.

Open-source models changed the math

ISNet and similar open-source segmentation models have closed most of the quality gap with commercial, closed-source alternatives over the past few years. That means the case for paying a per-image fee to a third party has gotten a lot weaker — the model doing the actual work can run on infrastructure you control instead of behind someone else's metered endpoint.

What this means for teams

• Predictable pricing that doesn't move with a busy week or a big batch job
• No rate limits tied to a billing tier — usage is a capacity question, not a spend question
• Images never leave infrastructure we operate, which matters for teams with data-handling requirements

Running the model ourselves is more work upfront than calling someone else's API. But it's the only way to offer background removal without asking every team to do mental math about their bill every time usage goes up — which is the whole reason Fnora exists.