With your current 8CPU setup, you can handle about 8-10 users processing images at the exact same second. Since background removal is CPU-heavy, each request usually claims a full core to keep the processing fast. You can likely have around 10 active users hitting the 'remove' button at the exact same time before things start to queue up.

You can use load testing tools such as Artillery and Locust to test your app. From my experience, a Locust cluster (1 master, 4 worker) deployed in Railway (32vCPU, 32GB RAM, 4 worker instances) can dish out around 100k RPS without issues (tested with 4 threads per worker instance). Going further introduces python SSL errors for some reason.