🧠 Why this game matters

Image generators (Stable Diffusion, DALL·E, Midjourney) model the statistics of photos, not the physics of cameras. Their tells live exactly where a CV engineer looks: noise floors, illumination consistency, perspective geometry, boundary artifacts. Deepfake detectors are trained on these same signals — you just ran a human version of one. As generators improve, forensic detectors chase them; it's an arms race you're now part of.