Generate App Store Screenshots with AI or Claude Code

Short answer

The safe way to generate App Store screenshots with AI is to capture your real UI first, then use AI for headlines, layouts, and variations around it. Capture with the Simulator or fastlane snapshot, have Claude Code write the copy and framing config, and check the set against Apple’s guideline 2.3.3. Don’t let an image model draw your interface.

OpenASO publishes this guide. It does not generate screenshots, though its MCP server can export competitor screenshots for review. We have not run the third-party tools below. Their details come from their repos, checked November 1, 2026.

What does Apple allow?

Apple’s guideline 2.3.3 opens with “Screenshots should show the app in use” (App Review Guidelines, updated October 16, 2026). A login page, title art, or splash screen isn’t enough on its own. Apple’s product page advice also says to use images captured from your UI. The guidelines don’t mention AI imagery, so assume the same rules apply however an image was made.

Apple’s guidance on 2.3.3 gives the detail:

• More than half must show the app in use. App Review judges the set for each device type. Three screenshots need two in use, and four need three.
• Other slots can be marketing. The remainder can use marketing material to show benefits and features.
• Overlays are fine. Text and image overlays are allowed if the app’s main functionality isn’t obscured.
• Don’t misrepresent. Don’t misrepresent the actual interface, include misleading marketing, or let visuals crowd out the app. Update screenshots with each new version so they match the current UI.

Other rules matter too. Section 2.3 says screenshots must accurately reflect the app’s core experience. Guideline 2.3.2 says to make clear when featured items need extra purchases, 2.3.9 asks for fictional account data rather than a real person’s, and 2.3.4 limits previews to screen captures of the app. Sizes are covered in App Store screenshot sizes.

What does a practical workflow look like?

Keep three layers separate: real captures, copy, and layout. Then a headline change never touches your UI.

1. Capture real UI

For a few screens, use the Simulator. Launch your app, then run this from Terminal:

xcrun simctl io booted screenshot home.png

Apple documents this command, and Simulator’s File menu option, in an older guide it has since retired, so menu names may have changed.

For many devices or languages, use fastlane snapshot, which runs your UI tests on each device and language combination:

fastlane snapshot init
fastlane snapshot

Add SnapshotHelper.swift to your UI test target, call setupSnapshot(app) before launch, and call snapshot("01Home") at each screen. In your Snapfile, list devices and languages. Set override_status_bar(true) for a clean status bar, which fastlane sets to 9:41 with full battery and signal.

Claude Code is useful here because it can write those UI tests and seed demo data. Use realistic, fictional content, and put the app in a state that shows real features.

2. Generate copy with AI

Give Claude your feature list and the real captures, then ask for several headline sets. Require that each headline describes something the capture shows. If you connect the OpenASO MCP, it can export competitor screenshots, so Claude can compare their first-screen hooks with yours. Skip claims you can’t support. Guideline 2.3.7 bars unverifiable claims in subtitles, and the same care suits captions.

3. Build layouts around the capture

fastlane frameit adds device frames, a background, and text. Claude can write Framefile.json, plus title.strings and keyword.strings for each language. background is mandatory in that file, and fastlane notes that output without titles keeps full resolution, so it can’t be uploaded directly.

Alternatively, ask Claude for an HTML template that places the unmodified PNG inside a frame, then export it at the exact sizes you need. Either way, make variations by changing copy, color, and order, not by editing the capture.

4. Check the set against Apple’s rules

Before you upload, go through each device set:

• Compare every screen with the current build, pixel by pixel where you can.
• Count the screens that show the app in use, and confirm it’s more than half.
• Confirm that overlays don’t hide the main function.
• Remove anything the app can’t do.

5. Upload and preview

Upload through App Store Connect or with fastlane’s deliver. Apple added an Asset Library and a preview tool in October 2026, and API 4.5.1 deprecates some older media resources, so confirm that your upload tool is current.

Which Claude skills and tools can help?

Several real options exist, and they differ mainly in whether AI touches the pixels.

rorkai’s asc-shots-pipeline skill covers capture, framing, and upload through the asc CLI, and calls its local screenshot commands experimental. For planning only, the skills we compared from Appeeky and ASO.dev write briefs rather than images.

Read any skill before you install it. The image-model route also sends your screenshots to another AI provider.

How can AI imagery go wrong?

The main risk is a screenshot that shows something your app doesn’t have. Image models return new pixels, so text can shift, buttons can change, and invented features can appear. Both image-model skills above tell the model to keep the real app screen, but an instruction isn’t a guarantee.

Watch for these problems:

• Redrawn screens that no longer match your build.
• Fake results, reviews, or numbers inside the UI.
• Photos of people or devices you don’t have rights to use.
• Enlarged UI panels that crowd out the app.
• Slides that are all ad, with no app in use.

Guideline 2.3.1 says misleading marketing is grounds for removing your app and terminating your developer account. Keep the raw capture beside each final image, so you can compare them before you submit.

Next, capture five real screens from your current build. Write three headline sets with Claude, and check the sizes in our screenshot guide before you export.