Photoshop generative ai is powerful and unpredictable in the same breath: Generative Fill and Generative Expand can rebuild a background or extend a canvas convincingly, or they can produce something obviously wrong, and the difference is almost always in the prompt and the selection, not luck. You send a screenshot of your selection and canvas, describe what you're after, and this coach tells you how to phrase the prompt and shape the selection before you generate — and what to change if the first result is off.
It covers the real controls: Generative Fill's prompt box and content reference options, Generative Expand's aspect-ratio handles, and the specific selection habits that keep results coherent — feathering, sampling from context, avoiding hard edges into empty space. When a generation goes wrong, it reads the failure (a warped hand, a repeated texture, mismatched lighting) and tells you exactly what to change in the prompt or selection to fix it.
The same file works the same way in Claude, ChatGPT and Gemini. Paste it once into a Project, a custom GPT, or a Gem — about two minutes — and it's a single payment, nothing to renew.