# I made an AI build me an App Store video factory. It made 465 videos. Then Apple rejected them.

> Source: <https://dev.to/kei24/i-made-an-ai-build-me-an-app-store-video-factory-it-made-465-videos-then-apple-rejected-them-120a>
> Published: 2026-08-13 09:21:27+00:00

Hi. It's Kei again.

Quick recap for new faces: I'm a solo iOS developer in Japan, I make a lot of small apps, and almost nobody downloads them. I write about it here because it's cheaper than therapy and the comment section is nicer than my download graphs.

Last time I mentioned, almost in passing, that I'd rebuilt the store media for a pile of my apps. A few of you asked how. This is the how. And it's a stranger story than I expected, because the part that broke wasn't the AI — it was Apple.

Let me be precise about what I did, because the internet is currently full of "I generated a video with AI!" posts and I don't want to be one of the dishonest ones.

I did not type "make me a beautiful app trailer" into some model and get a finished video. There's no AI-generated footage in these. No AI narration. No AI-written marketing copy. No AI music.

If you're here for magic pixels, I'm sorry. It's more boring and, I think, more useful than that.

I have a lot of apps. Each one needs an App Store preview video, in a lot of languages. Doing that by hand — record the screen, pick the moments, add captions, translate the captions, export at the right size, upload — is maybe an hour per video if you're fast. Multiply that by 20 apps and up to 35 languages each and you get a number that made me want to lie down on the floor.

So instead of making the videos, I made a **machine that makes the videos**, and I built that machine by pairing with an AI coding agent (I use Claude Code) for basically the entire thing.

Here's the honest division of labor:

So the accurate headline isn't "AI made my videos." It's "**AI built the factory, and the factory stamps out real videos.**" Each one is 24 seconds: a 3-second hook with the app icon bouncing in, then full-screen screenshots with a slow Ken Burns zoom and a caption fading in over each one. No voiceover — every claim is burned into the text, so it works with the sound off, which is how App Store videos autoplay anyway.

The machine rendered **465 videos.** Twenty apps, up to 35 languages each, about 15 seconds of render time per video. I watched a progress bar for a while and felt like a factory owner. It was, briefly, glorious.

If you want to see what actually came out the other end, the full-screen final versions are live on the App Store right now — here's the [pet care one](https://apps.apple.com/app/id6759948538) and the [chinchilla one](https://apps.apple.com/app/id6756287573). Tap the preview on either page; that's a video the factory made.

Here is the part nobody warned me about, and the reason I'm writing this post instead of just tweeting a smug before/after.

My videos looked, I thought, professional. Each screenshot sat inside a little iPhone frame — you know the look, the phone bezel around the screen, very clean, very "real product." I was proud of it.

Apple rejected it. **Guideline 2.3.4.**

The rejection text, which I have now read enough times to recite in my sleep:

The app preview includes content that does not sufficiently show the app in use... Includes framing... Includes device images and/or device frames.

Turns out: you are **not allowed** to put a phone frame around your footage in an App Store *preview video*. The video has to show the app basically full-screen, as if you're holding the phone yourself. The cute device mockup — the exact thing I thought made it look polished — was the violation.

And here's the twist that got me a second time: **this rule is only for videos.** For static screenshots, device frames are totally fine, encouraged even. So my brain had filed "device frames = professional" from years of making screenshots, and I carried that habit straight into video, where it's forbidden. Rejected once (an alarm app). Then, because I apparently don't learn, rejected again on a different app — where I had to pull all 39 already-uploaded videos and resubmit.

In my old design, the actual app screen only took up **44% of the frame.** The rest was bezel and background. Apple was, annoyingly, completely right.

The fix was to rip out the phone frame entirely and let the real screenshot fill the whole screen. Simple idea. The details were a series of tiny, stupid, character-building problems:

`fastlane`

, the tool everyone uses, doesn't support them. So the agent and I had to reverse-engineer the private endpoints App Store Connect's own website uses, and write our own uploader. That's the kind of afternoon that ages you.**AI didn't save me from the hard part. It moved the hard part.** The agent wrote thousands of lines of rendering and upload code faster than I ever could, and that was genuinely transformative — I would not have attempted this by hand. But it can't read your mind about Apple's review guidelines, it can't see that your safety-check script silently does nothing, and it definitely can't tell you that the thing you're proud of is the thing that'll get you rejected. That judgment was still my job. The AI is a spectacular set of hands. The eyes are still yours.

**The taste and the taste-checking are the moat.** Anyone can now generate 465 of something. Whether those 465 things are *right* — non-rejectable, non-tofu, actually flattering to the app — is still human work, and it's most of the work.

**And no, my downloads did not explode.** I know that's the question. Better videos are table stakes, not a growth strategy. But my store pages finally look like they were made by someone who cares, which — after two years of shipping into the void — is a thing I can at least control.

Same as ever. If you got this far, thank you.

I'll be back sooner this time, probably to complain about something else I did to myself. Thanks for reading — it still surprises me every time someone does.

— Kei
