Show HN: Punge: Ondevice NSFW Image Detection Using YOLOv11n, CoreML, TensorFlow
May 19, 2025 - news.ycombinator.com
Punge is a mobile app designed for on-device NSFW image detection, utilizing a custom-trained YOLOv11n neural network. The app operates entirely on the user's device, ensuring privacy by not uploading any data or requiring user login. On iOS, it uses CoreML for inference, while Android relies on TensorFlow Lite. The app processes images quickly, taking about 10–20 milliseconds per image on an iPhone 15, and achieves approximately 90% confidence in identifying key NSFW categories.
The app was developed as a privacy-first solution to help users clean up their camera rolls before sharing or handing over their phones. It emphasizes local inference and privacy-preserving AI, making it a useful tool for those concerned about personal data security. Punge is available on both the App Store and Play Store, and additional information can be found on the developer's website.
Key takeaways:
Punge is a mobile app for on-device NSFW image detection using a custom-trained YOLOv11n neural network.
The app operates entirely on-device with no cloud processing, ensuring user privacy.
Inference is fast, taking about 10–20ms per image on an iPhone 15, and achieves around 90% confidence in key NSFW categories.
Punge is available on both iOS and Android, utilizing CoreML and TensorFlow Lite respectively.
