Useknockout: open-source background removal API

Useknockout is a lean, focused API that solves a common but overpriced problem: removing backgrounds from images. While the market is saturated with AI wrappers, Useknockout differentiates itself by leveraging BiRefNet—a state-of-the-art model that actually holds its own on DIS5K and COD benchmarks—and pairing it with Modal's scale-to-zero GPU infrastructure. The result is a service that is remarkably fast (~200ms warm latency) and orders of magnitude cheaper than the industry incumbents. From a product perspective, the API is surprisingly robust for a project that claims to have been built in 'a few hours.' Beyond simple transparency, it includes high-utility presets like 'studio-shot' for e-commerce and 'sticker' for social apps. The inclusion of closed-form foreground matting (via pymatting) is a critical technical detail; it prevents the dreaded 'halo' effect and color spill, which is where most low-cost background removers fail. The real strength here is the transparency and flexibility. By being MIT licensed and providing a one-command self-hosting path via Modal, the developers are targeting the 'builder' crowd rather than just chasing SaaS subscriptions. You aren't just buying an API key; you're getting a reference architecture for serving heavy ML models efficiently. If there is a weakness, it is the 'beta' nature of the hosted version and a heavy reliance on Modal for the self-hosting experience. However, for any developer or e-commerce operation currently paying exorbitant per-image fees for simple segmentation, Useknockout is a no-brainer upgrade.