Ente Photos introduces encrypted library sharing and faster on-device machine learning

Ente Photos is rolling out two significant updates: a new library sharing feature that lets users share their entire photo collection with a family member in one step, and a machine learning overhaul that the company says makes on-device photo indexing up to ten times faster by moving its ML models onto the GPU.

Library sharing addresses one of the most frequent user requests: sharing a whole library with a partner, similar to Google Photos' Partner Sharing or Apple's iCloud Shared Photo Library, but with a difference Ente is keen to highlight: everything remains end-to-end encrypted by default. Users can share all albums they own at once via the settings, with the option to automatically share newly created albums going forward. Technically, each album keeps its own encryption key; when sharing is enabled, the user's device seals each key to the recipient's public key, so Ente's servers can establish the shares without ever being able to decrypt the photos themselves. The feature is initially available on iOS and Android for family plans, which allow up to five family members at no extra cost, with web and desktop support on the way. Notably, the feature is open source and available to self-hosters as well.

The second update targets speed. Ente's machine learning, which powers features like face recognition and search, and runs entirely on users' devices since the company's zero-knowledge design prevents it from processing photos server-side, now executes on the GPU for the first time. On a recent iPhone, running all models together takes roughly 40 milliseconds per photo, a tenfold improvement over the previous CPU-bound implementation. Rather than simply flipping a switch, the team rewrote portions of its models so that operations unsupported by GPU accelerators like CoreML and WebGPU (such as PReLU and expanded GELU activations) were replaced with mathematically equivalent fused operators, avoiding costly fallbacks to the CPU. Complementary optimizations in Ente's Rust pipeline, including a faster HEIC decoder and zero-copy image buffers, cut pre- and post-processing to just a few milliseconds per photo. The same accelerated pipeline now runs identically on mobile and desktop, with GPU support on macOS, Windows, and Linux.