Show HN: Punge: Ondevice NSFW Image Detection Using YOLOv11n, CoreML, TensorFlow
I built a mobile app called Punge that scans your camera roll for NSFW images—completely on-device. No cloud processing, no login, and no personal data ever leaves your phone.
Under the hood:
* It's powered by a custom-trained YOLOv11n neural network.
* On iOS, all inference runs through CoreML; Android uses TensorFlow Lite.
* Processing is fast—about 10–20ms per image on an iPhone 15.
* Achieves ~90% confidence in key NSFW categories.
* No data is uploaded, and the app doesn’t collect any user info.
This started as a privacy-first project to help people clean up their phones before sharing or handing them over. (You’d be surprised what gets forgotten in a camera roll.)
If you’re interested in privacy-preserving AI, local inference, or just curious how well this works in practice, I’d love your feedback.
iOS: App Store Android: Play Store Demo video: YouTube More info: https://markatlarge.com/