The hard problem in AI images was never generation. It was keeping things the same. Every model until now ran each edit as a full reroll. Ask for a smaller logo and the face drifted, the lighting shifted, the product warped. You gambled the 95% you loved to fix the 5% you didn't. GPT-Image 2.5 is the first one where I tell it what's frozen and it complies. That unlocks actual production work. 12 use cases with prompts you can paste straight in. 1. Lock a character "treat this image as the…