Photonarium is a photo and video catalogue that stays on your computer. It's for people who want the convenience of modern search and face grouping, without uploading their life to someone else's servers.
Screen guides:
- Gallery - browsing, sorting, info panel
- Full-screen viewer, Slideshow & Face tagging - viewing, playback, tagging
- Search - text search, filters, people, metadata
- Videos - video grid, scene timeline, heatmap
- Groups & Smart Groups - duplicates, directories, albums, saved searches
- Faces - face detection, people, recognition, Quick Match
- Database - folders, import, processing, settings
Installation & configuration:
- Installation - Docker, direct install, running, configuration
Other:
Most photo apps push you towards the cloud. That is great until you care about privacy, subscriptions, slow uploads, or working offline.
Photonarium keeps your library local and helps you do the three things people actually want:
- Find photos and videos quickly, even when you cannot remember filenames, and exclude what you don't want
- Tidy a messy collection, especially duplicates and near-duplicates
- Organise around people, favourites, and your own notes
Find out more about the motivations behind Photonarium in BACKGROUND.md.
- AI-powered search that understands what you type (e.g. "sunset over mountains", "birthday cake"), with negative terms to exclude concepts (e.g. "beach -people"). Named people in your search text are detected automatically. Wildcard date filtering lets you search for patterns like "every May 14th" or "October through February, any year". Search images, videos, or both - video search finds matching scenes within clips, with a heatmap showing match strength across the timeline.
- Face recognition - automatic face detection and recognition. Name faces and Photonarium finds them across your library. Easily pick a preferred face (avatar) for each person, and ignore faces of strangers.
- Photos and videos - manages images and video files side by side. Videos are broken into scenes with keyframe thumbnails, and can be searched at the scene level.
- Video transcriptions - automatic speech-to-text for video audio, enabled by default with automatic language detection. Transcription text is displayed as subtitles during playback, shown in scene preview popups, and is semantically searchable. Edit or add subtitles inline on the Videos screen, and set each video's language individually for accurate transcription of multilingual libraries.
- Duplicate detection at four levels of similarity (identical, near-identical, similar, related) plus auto-generated directory groups and user-curated custom groups (albums), with Refine Groups to filter by quality and view the best (or worst) images in the Gallery, or prune duplicates to trash
- 100% free and open source - Apache 2.0 licensed, no cloud, no accounts, no tracking. Runs entirely on your machine; your files never leave your computer.
- Automatic captioning - generate image descriptions using AI (BLIP/BLIP-2), then edit them if needed. Captions are searchable.
- Quality scoring - AI aesthetic scoring ranks your images by visual quality. Find your best shots instantly.
- Sort by similarity - select any image or video and sort the entire collection by visual similarity to it. Great for finding related shots.
- Smart Groups - saved searches that stay up to date automatically. Set your filter criteria once and matching items appear whenever you open the group, even files added later.
- Camera RAW support - native support for 20+ camera RAW formats alongside JPEG, PNG, and other standard image types. No conversion needed.
- Multi-device sync - use Photonarium from multiple devices at once. Changes made on one (naming faces, rating photos) appear on all others within seconds.
- Setup assistant - a guided first-run flow that configures Photonarium for your hardware and preferred search language, then downloads the required AI models. Choose from hardware presets (NAS, laptop, desktop) and language options (English or multilingual) without touching config files.
- Camera data and histogram - full EXIF metadata extraction with image histogram display. Search and filter by camera, lens, ISO, aperture, shutter speed, and more.
- Slideshow mode in the full-screen viewer with smooth cross-fade transitions, linear or shuffled playback, configurable timing, and video autoplay
Once you have run the model downloader, the models stay on your machine. Everything runs locally.
- Start the Photonarium app in a terminal window.
- Open the Photonarium web page in your browser, the default is http://localhost:5000
- On first launch, the Setup Assistant walks you through choosing a hardware profile, search language, and downloading AI models — all in a few clicks.
- Go to Database and add one or more folders that contain photos and videos.
- Let it scan. Big libraries take time, especially face detection and video processing.
- Go to Gallery and start browsing.
- Use Search when you want to find specific images or video scenes.
- Use Videos to browse your video library and explore scene timelines.
- Use Groups when you want to clean up duplicates or organise items into albums.
- Use Faces when you want to name people and improve recognition.
Photonarium is currently in beta. It works, and people are using it day-to-day, but installation is still a manual process, the interface is evolving, and you may encounter rough edges. If something breaks or feels wrong, please open an issue - feedback during this stage is especially valuable.
Photonarium is a desktop application that runs in your browser, not a mobile app. The backend (Python) and the frontend (the browser tab) are designed to run on the same machine - typically your laptop, desktop PC, or a home server where your photos and videos are stored.
It also works over a local network: you can run the backend on one machine (say, a NAS or always-on PC) and open the UI in a browser on another. Most features work fine in this setup, with a few caveats:
- Performance - it may be a little slower, for two reasons: 1) the backend machine needs some grunt, ideally a decent GPU, especially during image and video ingestion, and 2) quite a bit of data between the backend and UI needs to go over your local network.
- Reveal in folder opens a file-manager window on the machine running the backend, which is only useful if that's also the machine you're sitting at. On a headless server this will either silently fail or pop a window nobody sees.
- The folder picker (for adding directories) likewise opens on the backend machine. If you're accessing Photonarium remotely, use the CLI instead:
python app/app.py --add-folder /path/to/photos.
The UI adjusts for smaller screens and touch input - the toolbar collapses to a hamburger menu, touch-friendly scroll zones appear, and layout stacks vertically. It's usable for browsing and basic tasks on a phone or tablet, but the full experience (drag-box selection, keyboard shortcuts, side-by-side info panel) is designed for a desktop browser. Think of mobile as a handy way to flick through your library on the sofa, not a replacement for the desktop workflow.
Photonarium has no user accounts, no login, and no access control. Anyone who can reach the server's address can view and modify your library. This is fine for personal use and trusted home networks, but you should not expose it to the public internet.
Multiple browser tabs or devices on the same network can use Photonarium at the same time. Changes made on one client, naming a face, rating an image, creating a group, trashing a photo, are automatically pushed to every other open client within a couple of seconds. If a client falls too far behind (e.g. a laptop lid was closed for a while), it detects the gap and silently reloads to catch up. If your browser loses the connection with the Photonarium backend for any reason, your changes are blocked with a warning message until the connection is restored.
Use the toolbar buttons, or these shortcuts (ignored while you are typing in a text box):
- Ctrl/Cmd + G: Gallery
- Ctrl/Cmd + M: Manage Database
- Ctrl/Cmd + D: Groups (Duplicates and Albums)
- Ctrl/Cmd + S: Search and Filter
- Ctrl/Cmd + F: Faces and People
Common keys across screens:
- Escape: go back / close a panel (for example: exit Search, Duplicates, or Database back to Gallery; close dialogs; close full-screen).
- Enter: open the selected item (where it makes sense, like opening an image or video).
- Delete / Backspace: remove selected items (where supported, usually with a confirmation).
Most screens use the same selection behaviour:
Mouse and trackpad:
- Click to select.
- Ctrl/Cmd + click toggles an item in the selection.
- Shift + click selects a range (from the last "anchor" selection).
- Right click toggles an item in the selection.
- Drag on empty space to draw a selection box:
- Left button: replaces the selection with what is inside the box
- Right button: toggles everything inside the box
Touch:
- Tap to select.
- Long press to add to the selection without clearing.
Keyboard (in any grid view):
- Arrow keys move the active selection.
- Shift + arrows extends the selection.
- Page Up / Page Down moves by a page.
- Ctrl/Cmd + Up / Down jumps to first or last item.
- Ctrl/Cmd + A selects all.
- Escape clears the selection.
- Large imports and database rescans take time. Let it run and come back later.
- Face recognition improves as you tag more clear examples of the same person.
- Add multiple people before refining any one person, this tends to reduce false matches.
- Once you have several people established, use Quick Match (sparkle button) to rapidly identify unknown faces. It shows you the most likely matches based on face similarity.
- If two people get mixed up, increase that person's recognition threshold.
- Emoji ratings work well for quick favourites, and make filtering pleasant.
- Use negative terms in search (
beach -people) to exclude concepts from results. - When searching for video content, use "Videos" mode to see scene-level results with a heatmap showing exactly where your query matches within each clip.
Photonarium is built on the shoulders of some remarkable open-source AI/ML work:
- OpenCLIP (LAION) - the semantic image embeddings that power search and similarity
- BLIP / BLIP-2 (Salesforce Research) - automatic image captioning
- facenet-pytorch - MTCNN face detection and InceptionResnetV1 face recognition
- LAION Aesthetic Predictor (LAION) - lightweight aesthetic quality scoring
- NIMA - neural image quality assessment trained on human aesthetic judgements
- PyTorch (Meta) - the foundation all of the above is built on
Thanks to the broader Python community - Flask, Pillow, NumPy, OpenCV, and countless other libraries - for making a project like this feasible for a small team.
The tutorial example images come from Lorem Picsum, which serves freely usable photos from Unsplash and the example videos were created using downloads from pexels:
- Flowers
- Cars
- Football
Finally, thanks to Anthropic and Claude Code for doing a lot of the grunt work.
We hope you enjoy Photonarium and find it valuable. If you'd like to show your support, please use one of the links below:
Copyright (c) 2026 7th software Ltd. - Licensed under Apache 2.0