# Building an AI-Powered Photo Cleaner: Lessons from the App Store > Source: > Published: 2026-07-04 04:17:26+00:00 Last year my iPhone popped the dreaded "Storage Full" notification for the hundredth time. I checked the breakdown: 68% was photos. Not memories — junk. Burst shots with 20 variants of the same frame. Screenshots from 2023. Duplicates from a failed iCloud sync. I couldn"t find a cleaner that worked the way I wanted, so I built one. Here"s what I learned shipping a SwiftUI + Core ML photo management app to the App Store. Photo cleanup apps aren"t new. But most of them are either too manual (swipe thousands of photos one by one) or too aggressive (auto-delete without context). The hard part isn"t feature detection — it"s the UX of trust. Users need to feel in control. Every deletion should be their decision. The app"s job is to *surface* what to delete, not to delete it for them. Using Core ML"s Vision framework, we compute perceptual hashes (pHash) for every photo in the library. On-device processing means nothing leaves the device: ``` js let request = VNGenerateImageFeaturePrintRequest() let handler = VNImageRequestHandler(cgImage: ciImage) try handler.perform([request]) ``` Similar photos (hash distance < threshold) are grouped. The algorithm naturally catches burst photos, near-duplicate screenshots, and re-imported images. This was the hardest feature. The app scans for text overlays and document-like patterns (passports, IDs, tax forms) using a combination of Vision text recognition and heuristics. All processing stays on-device — no network requests for image analysis. HEVC re-encoding can reduce video storage by 60-80% with minimal quality loss. The trick is doing it in the background without the user noticing. We A/B tested 4 sets of App Store screenshots. The winner (showing before/after storage numbers) converted 3x better than the runner-up (showing app features). People scroll screenshots looking for proof, not feature lists. "Photo cleaner" has 10x the search volume of "duplicate photo remover". We optimized for the broader term and caught both audiences. The first 20 reviews came from TestFlight beta users. Having reviews on day 1 made a huge difference for ranking. Six months in: steady organic downloads from App Store search, high retention (users clean in bursts), and the best feedback: people telling us they finally freed up 30GB+ of storage. If you"re building a utility app for iOS, the biggest takeaway is this: **on-device AI + clear user control = trust**. And trust is the only currency that matters for an app that touches your photos. *Swipe Cleaner is available on the App Store. Built with SwiftUI, Core ML, and a lot of late nights.*