PhotoPrism improves face recognition, adds Vulkan transcoding, and revamps info sidebar

PhotoPrism’s latest update fully transitions face recognition to an ONNX-based pipeline, replacing the legacy Pigo detector. This change allows users to choose mixed-case model names from Hugging Face, Ollama, and OpenAI-compatible catalogs, increasing flexibility for those using advanced artificial intelligence features.

In terms of user interface, a redesigned info sidebar streamlines metadata management by enabling both direct and dialog-based editing of metadata, albums, and labels. Following this, the viewer sidebar now mirrors the face management actions found in the Edit People menu and lets users manually mark or assign faces missed by automatic face detection.

Media handling advances now include video transcoding with Vulkan hardware acceleration via FFmpeg 8. A native HEIC/AVIF image reader is in place, bolstered by an upgraded libheif (version 1.21.2), while support for layered TIFF and Adobe Photoshop PSD files is now available.

To further enhance workflow, label filtering introduces NOT and AND operators, file uploads support drag-and-drop, and zstd compression is used to speed page loads. Backend improvements feature hardened WebDAV compatibility and a new Ubuntu 26.04 LTS base image, improving deployment reliability.

Additionally, missing interface translations were generated with DeepL and Google Translate, inviting native speakers to refine them as needed.