Managing a massive Apple Photos library is a losing battle for most of us. You spend hours scrolling through blurry screenshots, accidental pocket videos, and near-identical bursts of the same sunset, but the sheer volume makes manual cleaning impossible. I’ve been looking into Piqt, a new tool that tries to solve this using on-device machine learning rather than sending your private data to some massive LLM cluster.
If you are interested in the prompt engineering or the underlying vision models, the developer is actively looking for feedback on the specific research and techniques used. This is a much more interesting approach to an AI workflow than the standard "upload everything to our server" model. Most "smart" gallery apps feel invasive, but building a complete guide for on-device deployment like this is a much harder—and more ethical—technical challenge.
The developer actually started this project in Jupyter notebooks, experimenting with how much heavy lifting AI models could do for automation before moving it into a polished mobile deployment. It’s not just another "storage cleaner" that tries to sell you more iCloud space; the actual goal here is to make the library navigable again through intelligent ranking and sorting.
Here is the technical breakdown of how this approach works:
Local Execution: It relies on a combination of Apple’s Vision API, various ML models, and clustering algorithms. Everything stays on your iPhone.Privacy-First Architecture: There are no accounts to create and no cloud sync involved. The dev even implemented a validation step in the release process to ensure the app literally cannot call the iOS delete APIs. It stages "junk" in a separate album so you retain ultimate control.Algorithmic Curation: Instead of just looking for large files, it uses vision models to understand the content and rank what is actually worth keeping.
If you are interested in the prompt engineering or the underlying vision models, the developer is actively looking for feedback on the specific research and techniques used. This is a much more interesting approach to an AI workflow than the standard "upload everything to our server" model. Most "smart" gallery apps feel invasive, but building a complete guide for on-device deployment like this is a much harder—and more ethical—technical challenge.
If you want to see the specific technical decisions and the design philosophy behind the model selection, you can check out the developer's deep dive here:
https://piqt.app/blog/piqt-design-values/
And the app itself is available on the App Store:
https://apps.apple.com/us/app/piqt-photo-video-curator/id6778080898
Story tracker · related coverage
Apple users in high-risk roles should probably be checking their 15d ago
Apple is building its own AI model for China with Alibaba's help 15d ago
Apple is finally building its own LLM for China with Alibaba's 16d ago
Apple is building a custom LLM for China using Alibaba's 16d ago
Using BLE to play poker on flights is a clever workaround 23d ago
Next Shaming people for sharing AI slop won't stop the flood of →
Free AI toolbox — all free to use
All Replies （1） #
C
Sounds like more marketing hype. Unless it actually cleans my junk files, I’m not interested. Show some proof.
0