Three months after its December update, PhotoPrism, a self-hosted open-source photo management tool, has released a new version featuring improved Ollama-based AI models and support for configuration via the OLLAMA_BASE_URL and OLLAMA_API_KEY environment variables.
Moreover, the app now includes a fallback mechanism for caption generation when reasoning models return “thinking” responses instead of direct captions.
Several fixes address edge cases in indexing and media processing. A bug causing merged photos to retain incorrect image types after merging with video files has been resolved. The update also corrects minor GPS coordinate overshoots near map boundaries in the Places view.
Folder-based albums have received multiple fixes. Conflicts involving emoji-named subfolders and parent albums have been resolved, and path collisions from truncated folder slugs have been addressed. The interface now provides clearer feedback by displaying file error reasons directly in both Card and List views.
Improvements also cover thumbnail generation and metadata handling. Error buffer handling has been corrected when the interoperability index is missing, and invalid GPS coordinates from Google JSON metadata are now clamped to valid geographic bounds to prevent mapping issues.
Command-line behavior and authentication compatibility have been refined as well. Input path validation and exit codes are now standardized, and there is enhanced support for the Google OpenID Connect identity provider when using SQLite.
On the server side, the update adds new HTTP security configuration options and improves the configuration and performance of Gzip route exclusions. Additional fixes address logging behavior and media processing. The system now properly handles missing caption thumbnails and errors during video remux operations.
For detailed information, visit the changelog.
