RealTime AI Cam 1.3.6 (iOS, in App Store review)

iOS 1.3.6 (build 44): submitted to App Store review on Oct 29, 2026. Android 1.23 carries the same model and comes next.

• What's this? now runs LFM3-Omni-2B by Liquid AI on the phone through llama.cpp (735 MB, still fully offline). On Matt's 20 real photos it scored about 10 to 13 points higher than the old Vision Narrator 0.8B on both phones.
• More reliable when you switch apps or lock the phone mid answer: the app waits to be back in front, reloads the model once if the GPU refused, and never goes silent.
• One answer at a time: a second request (like asking a question while it is still describing) now waits for the first instead of crashing the app.
• The model credit and the LFM Open License v1.0 are in Settings under Credits, and in CREDITS.md and NOTICE.md.

RealTime AI Cam 1.2.1 (iOS) / 1.6 (Android)

The version that is live in the stores right now: iOS 1.2.1 (build 28) on the App Store and Android 1.6 (versionCode 11) on Google Play.

A camera that tells you what it sees, out loud and offline. Nothing leaves the phone.

• Object detection with YOLOv8, 601 classes from Open Images V7, with indoor and outdoor modes
• Scene and page description from the on-device narrator
• Live English OCR read aloud, plus a mail and bill mode (what it is, who sent it, the amount, the deadline)
• Offline Spanish to English for signs and labels
• LiDAR distance per object on Pro iPhones
• Built for VoiceOver and TalkBack

Fixed in 1.2.1: the crash when opening the Info Guide, and the flashlight now turns off when you leave the app or it goes to the background. UI tests cover first launch, detection, reading and the guide.

Get it: App Store · Google Play. Source for both platforms is in ios/ and android/.