Generate App Store Screenshots with Gemini CLI

Gemini CLI Meets App Store Screenshots: Why Bother?

If you spend hours dragging frames in Figma or wrestling with Photoshop’s alignment tools just to hit yet another App Store Connect submission, you’re not alone. Most indie teams fall into two camps: those doing everything manual, and those automating the boring stuff because they have releases to ship and bug lists longer than Tim Cook’s keynote script.

We’ve already seen what tools like GitHub Copilot or Cursor can automate for code, but what about Apple’s most tedious requirement: the full set of screenshots in every required pixel size? This is where the AI-agent tool craze promises big things. So I ran Gemini CLI—Google’s latest command-line interface for Gemini AI—through a reliable series of real-world App Store screenshot workflows to see if it lives up to the hype.

First: What Is Gemini CLI (and Is It Ready for Designers?)

Gemini is Google’s LLM, but the CLI gives you terminal superpowers. Install via pip install google-generativeai (assuming your Python setup isn’t broken). You authenticate, drop into the command line, and prompt it like you would in their web interface—but now, system prompts and scripts are on your turf. For code, documentation, even content creation, it’s solid. For visuals, it’s chasing—but catching up.

You can feed Gemini CLI system prompts ("Compose a series of PNGs in iPhone 6.7-inch format with these features: …") and pipe outputs straight to files. In practice, this is kludgier than Midjourney or DALL·E, but infinitely more scriptable. The question: does it save time, or just generate more work?

What Actually Works with Gemini CLI for Screenshot Generation

1. Asset Generation (with a Catch). If you prompt Gemini CLI for "App Store screenshots for a productivity timer app (5 screens, English, 1290x2796, clean device frame, iOS 17 UI)", you’ll get URLs or base64 blobs, often PNG. Sometimes you land mockups that look halfway decent—headline, background, fake UI. Sometimes you get AI hallucinations: calendar UIs that exist nowhere, fonts Apple has never shipped. Expect a 60% hit rate if you’re specific, and zero on anything offbeat (try requesting an Apple Watch frame—good luck).
2. Batching Flows. The superpower here is running prompts in a loop—pass in your translated headlines or screenshot copy, and pipe assets into a folder for your next build. A simple bash script can batch up all store sizes, especially if you pair with an upscaler or a sharp image processor.
3. Variants for Experiments. Need A/B test assets for App Store Connect Product Page Optimization? Gemini CLI can hack out headline-messaging variants faster than any human designer you’d pay. Copy, frame, color schemes, backgrounds—not the world’s best, but you’ll discover what doesn’t work fast.

What Gemini CLI Gets Wrong (and Where It Gets Ugly)

Quality is erratic. Sometimes, the screenshots look like decent Store shots for a generic stock app. Other times, the iPhone bezel isn’t right, the corner radius is off, or there’s too much space wasted on fake device chrome.

But this is where it gets messier. If you push Gemini to generate device-accurate UI—say, iOS 17’s sheet views, the latest tab bar, Cupertino controls—it often makes subtle but damning mistakes. That’s enough for Apple to reject your submission if a reviewer catches it.

Localization is another sticking point. Gemini CLI does fine with headlines—"Track your time with zero effort," etc.—but don’t trust it to get non-English copy pixel-perfect, especially for RTL scripts. A run through an AI translator mixed with human review is still required. I don’t trust Gemini for Japanese or Arabic layout, and neither should you if you want to pass manual review.

One more: Gemini can’t (yet) overlay live device screenshots from your app. You’ll either hand it static images from your latest build, or have it invent something based on a prompt. In theory, someone could script a flow where Gemini applies labels and backgrounds to real screenshots, but right now that glue code is DIY, and quality is inconsistent compared to dedicated tools.

Putting Gemini CLI to Work—A Concrete Example

Suppose you’re running a solo shop. Your main pain is the annual iOS device size shuffle and churning out consistent screenshots for your simple to-do app. Here’s how you could script it:

for size in "1290x2796" "1242x2688" "1170x2532"
do
  gemini "Create an iOS 17 App Store screenshot for a productivity to-do app. Size: $size. Device frame, modern UI, headline: 'Focus on your day.' Minimalist background." \
  --output screenshots/$size-1.png
done

That batch loop kicks out three device sizes, one screen each. You’ll need some cleanup—sometimes Gemini produces goofy text styles or weird margins. But this beats slicing Photoshop layers by hand.

If you want to go further (headline translations across locales, five screens per device, color variants for A/B), just extend that loop, vary the input text, and keep a human in the loop to check for fails. I’d never trust any pure-AI run for a release. But it gets you most of the way there.

When to Use Gemini CLI—and When to Look Elsewhere

• Speeding up batch-E2E flows. Prototype, experiment, blast out early variants, or speed up hackathon releases—Gemini CLI fits right in. If you’re prepping screenshots for an MVP, an investor deck, or your first pre-launch TestFlight, give it a try.
• Losing patience fiddling with layouts. If you care about speed (and don’t care if a corner radius is one pixel off), Gemini CLI stops the busywork. This is not for boutique launch visuals.
• Not great for detail-critical, production assets. If you’re aiming for a polished, App Store-ready set of screenshots, or hitting Apple’s accessibility specs, you want control Gemini can’t give. For real production, I recommend dedicated visual tools like ScreenshotWhale (especially for guaranteed layout accuracy, automation across locales, and layering actual device screenshots where needed).

Don’t treat generative AI as a silver bullet. The “magic” is more of a productivity multiplier for prototypes, not a handoff for your final store visuals. If you’re fastidious about your App Store presence (and you should be, if you’re serious about downloads), always layer human QA—or swap in tools built for this very task, with concrete documentation on Apple’s and Google’s visual requirements.

Tips for Getting the Most from Gemini CLI in Screenshot Generation

• Get ruthless with prompts. Spell out every requirement: dimensions, device frame specifics, amount of copy, visual style, background color. Details matter. AI shorthands and vague requests multiply errors.
• Check every output pixel. Device frames, navigation bars, safe zones, and text contrast can get wonky. Stack the AI’s output against Apple’s App Store screenshot guidelines and the official device templates.
• Combine output with scripts. Use imagemagick or sharp to resize, trim excess whitespace, and add overlays with real device screenshots from your app. AI visuals work best when brought into a more predictable tooling chain.

Loose Ends and Where Gemini CLI Fits

As of now, Gemini CLI is less a designer replacement and more a brute-force shotgun for batch asset generation. The real productivity comes in getting 80% of your screenshots prepped in bulk, then fixing the remaining 20% by hand or by using a specialized SaaS. If you’re shipping fast and want to avoid jumping between Canva and the shell 20 times per release, Gemini CLI is worth the test. For launch campaigns or client work, don’t skip pro QA—or delegate to a tool built for production, like ScreenshotWhale, which is built for actual App Store asset requirements (full disclosure: no AI hallucinations, no guessing corner radii).

We haven’t tested Gemini CLI outputs at scale for tablets or Apple Watch assets—early runs failed to generate device-accurate frames reliably. For now, stick to iOS and Android phone mockups, and apply a heavy dose of skepticism before upload.