Five images, going up on the App Store listing: Home, saving from another app, search, locked items, folders. Each one is a real screen from the real app, framed on a phone, with one line of headline copy over a gradient background. They look simple. Getting there wasn't.
The rule that shaped everything
App Store screenshots have to accurately represent the actual app — that's an explicit review guideline, not a style preference. So the phone screen in every one of these five images had to be a pixel-exact capture of Snugit, not an approximation of what Snugit roughly looks like. Everything else — the background, the floating decoration, the headline — could be generated. The screen itself couldn't be faked, even a little.
Where AI actually helped, and where it didn't
Asking an image generator to invent the whole scene from a text description came back looking like a screenshot of an app that doesn't exist — close, wrong in the details, occasionally garbled text. What worked was flipping the task around: hand the model the real screenshot as an input image, and instruct it to build a frame and background around that exact image without altering it. Editing around a fixed, real anchor turned out to be a job these tools are actually good at. Inventing the anchor itself was not.
The order matters more than any individual image
The five screenshots aren't random — they follow the same sequence as the app's own onboarding: what Snugit is, how you save into it, how fast you find things again, then the privacy angle, with folders last since it's the smallest of the five ideas. Anyone who sees the screenshots first and opens the app later gets the same story told twice instead of two different pitches that don't quite agree with each other.
Download Snugit — you'll see the same five screens for real, not just in a listing.
More on how the first-run version of that same story got built: onboarding, finally.