한 명의 방문자가 GPU를 사용하지 않고 익명의 AI 사진 편집기 구축

I’m building Turner AI, a browser-based AI photo editor: https://turner.art

The visible workflow is simple: upload a photo, write what should change, and download the result.

The harder engineering problem is fairness.

I wanted the first edit to work without an account. But image editing is not a zero-cost request, and anonymous traffic means a small number of visitors can consume a disproportionate amount of GPU time.

The design we are moving toward separates convenience from authority:

• browser state can discourage accidental overuse;
• the server remains the source of truth for admission;
• usage is counted only after a job is actually created;
• queues and fair-use controls protect everyone else when capacity is tight.

That last point matters. A failed upload, a failed human check, or a rejected request should not silently spend someone’s allowance.

The product is free to try, requires no account, and successful downloads do not carry a Turner watermark.

For developers who have shipped anonymous compute-heavy tools: would you add signup earlier, or keep the first-use flow open and enforce fairness behind the scenes?