All the main features.
Every feature documented. No marketing promises.
Up and running in minutes. No terminal needed on Windows or Linux.
OffGallery Manager installs everything automatically: Miniconda, Python, libraries, AI models and optionally Ollama for LLM generation. A double-click on Windows, a graphical wizard on Linux.
OffGallerySetup.exe — guided installer
Download the executable from the Release page, double-click it and follow the 5-step wizard. The Manager detects your GPU (NVIDIA/AMD/CPU), selects the right dependencies automatically and installs everything without opening a terminal. When done, it creates a Desktop shortcut to launch OffGallery.
⬇ Download OffGallerySetup.exeIf SmartScreen shows a warning: click "More info" → "Run anyway". The warning is due to the absence of an EV code signing certificate, not a security issue.
OffGallerySetup — guided installer
Download the executable, make it runnable with chmod +x OffGallerySetup and launch it. The wizard installs Miniconda, creates the Python environment, installs libraries and sets up the application menu launcher. No terminal needed.
Tested on Ubuntu 22.04+, Fedora, Arch. Requires a graphical interface (X11 or Wayland).
Bash script — M1/M2/M3/M4 and Intel
Run bash installer/install_offgallery_mac_en.sh from the project folder. The script supports both Apple Silicon (Metal/MPS) and Intel. On completion it creates OffGallery.app in ~/Applications, searchable via Spotlight.
Lightweight, Full or Custom
The wizard offers three profiles: Lightweight (~14 GB — SigLIP, DINOv2, Aesthetic, BioCLIP, Argos, no Ollama), Full (~18 GB — everything + Ollama + qwen3-vl:8b) and Custom with a checkbox for each component. Automatic GPU detection: NVIDIA CUDA, AMD DirectML or CPU-only.
Local AI models, orchestrated in one system.
All models run locally. After the first download, no internet connection is needed.
SigLIP semantic search
Matches text queries against the visual content of every image. SigLIP is natively multilingual (400M parameters, trained on 10B pairs) — search in Italian, English or any language without translation. Search "heron in flight at sunset" without tags, without keywords.
google/siglip-so400m-patch14-384DINOv2 visual similarity
DINOv2 generates visual embeddings for similarity search: find images similar to a selected one and detect near-duplicates in the archive.
facebook/dinov2-baseBioCLIP species recognition
Recognizes approximately 450,000 biological species (birds, mammals, insects, plants, fungi) from the TreeOfLife database. Produces a 7-level taxonomic hierarchy visible in the tooltip of each photo. No other consumer cataloging software has this feature.
imageomics/bioclipLocal LLM Vision
Generates titles, descriptions and tags by reading the image. The model runs locally (Qwen3-VL-4B or larger via Ollama/LM Studio). If BioCLIP has identified a species, the scientific name is inserted automatically in the title and tags — the model doesn't invent it.
Ollama / LM StudioAesthetic scoring
Aesthetic score from 0 to 10 based on a scoring model trained on photographic datasets. Displayed in gallery as an overlay badge on each thumbnail.
Aesthetic PredictorTechnical quality MUSIQ
MUSIQ analyzes perceived technical quality: sharpness, noise, blur. Score separate from aesthetics — useful for quickly discarding technically failed shots.
MUSIQ / pyiqaCompatible with existing photographic workflows.
XMP, Lightroom, EXIF, offline geolocation. No disruption to your workflow.
Bidirectional XMP sidecar
Import and export metadata via standard XMP sidecars. Tags, ratings, descriptions and taxonomic hierarchies compatible with Lightroom, Capture One, digiKam and others. A badge on each thumbnail shows whether the sidecar is current, outdated or missing.
Lightroom catalog import
Reads .lrcat catalogs (SQLite) read-only. Imports the file list with ratings, flags and color labels directly from Lightroom Classic, without touching the original catalog.
Full EXIF extraction
ExifTool extracts over 200 EXIF/IPTC/XMP fields: camera, lens, ISO, aperture, speed, GPS data, copyright, flash. Searchable and filterable from the search tab.
Offline geolocation
GPS coordinates become a readable geographic hierarchy (Continent › Country › Region › City) without external APIs. GeoNames data (~145,000 cities) is bundled in the software. The result is searchable and filterable from the search tab.
Archive statistics
Stats tab with distribution by camera, lens, ISO, aperture, shutter speed, rating and color label. Quantitative overview of your photographic archive.
CSV export
Export metadata from the entire archive or a filtered selection to CSV for external analysis, readable backups or integration with other tools.
Find any photo in seconds.
Semantic search, fuzzy matching, visual similarity. 25+ RAW formats supported.
Combined semantic + tag search
Combine SigLIP semantic search with filters for tags, Lightroom ratings, color labels, camera, lens, date and geographic location. All filters applicable simultaneously.
Fuzzy tag matching
Tag search uses fuzzy matching: finds "heron" even if the tag written is "grey heron". Configurable tolerance to handle linguistic variants.
Visual similarity search
Select an image and search for visually similar ones via DINOv2 embeddings. Useful for finding variants of the same shot or near-duplicates before culling.
25+ RAW formats supported
Canon CR2/CR3, Nikon NEF/NRW, Sony ARW, Fujifilm RAF, Olympus ORF, Panasonic RW2, Pentax PEF/DNG, Leica DNG, Phase One IIQ and more. Plus JPG, PNG, TIFF, HEIC.
Multilingual interface
GUI available in Italian, English, French, German, Spanish and Portuguese. LLM output language configurable separately in any language supported by the local model.
Visual badges in gallery
Every thumbnail shows up to 6 overlay badges: aesthetic score (0-10), MUSIQ technical quality (0-100), Lightroom rating (★), color label, semantic similarity (in search results). A sixth badge indicates whether the XMP sidecar is current, outdated or missing — without opening any detail.
An architecture built to grow.
Add and switch LLM backends, integrate new data sources — without modifying the core application.
LLM Backend
Ollama LLM Vision
Connect OffGallery to a local Ollama server for natural language tag, title and description generation.
Communication via local REST API (port 11434). Auto-discovered at startup if the server is running.
LLM Backend
LM Studio
Use LM Studio as a local OpenAI-compatible server. Ideal for those who prefer a GUI for managing downloaded models and configuring quantization.
Compatible with any vision model loaded in LM Studio via local API (port 1234).
Plugin by Riccardo Merlotti
Free pluginNature photography
BioNomen
Enriches BioCLIP tags with common species names in multiple languages. A grey heron identified by BioCLIP automatically becomes "airone cenerino", "héron cendré", "Graureiher" — in XMP sidecars, offline.
Data comes from the GBIF database, bundled locally. No internet connection required after installation.
Essential for wildlife, botanical and nature photographers archiving large volumes of photos with biological subjects identified by BioCLIP.
Free pluginGeographic data
GeoNames
Downloads and indexes the GeoNames geographic database for selected countries. Search by name or derive a hierarchy from GPS coordinates with higher accuracy than the built-in engine.
Produced hierarchy: GeOFF|Continent|Country|Region|City. The leaf node is automatically added as a tag. Data is searchable and filterable from the Search Tab.
Essential for photographers shooting villages, hamlets and minor places not covered by the built-in engine (~145,000 cities). With GeoNames: millions of places.
Free pluginGeographic data
GeoSpecies
Refines BioCLIP classification using only the species expected at the photo's GPS location.
Narrows the field from ~450,000 global species to a relevant geographic subset, increasing taxonomic accuracy. Checklists downloaded from GBIF, then used offline.
Free pluginGeographic data
NaturArea
Identifies the protected area (WDPA/UNEP database) and habitat type (ESA WorldCover) for each photo with GPS coordinates. Data is written to the DB and XMP sidecars.
Recognised habitats: forest, shrubland, grassland, cropland, urban, wetlands, water, snow and more. Protected area searched via local WDPA database (~500 MB, downloaded once) — no external API.
Enables filtering photos by protected area or habitat type directly from the search panel.
Free pluginGeographic data
Weather Context
Retrieves historical weather conditions at the time of the shot via Open-Meteo: temperature, humidity, wind, precipitation and general condition (clear, rain, snow…).
Conditions are searchable from the Search Tab (e.g. find all photos taken in the rain) and visible on gallery image cards.
Requires an internet connection at runtime to query the Open-Meteo API (free, no key required). Local cache avoids repeated API calls.
Free pluginLLM Backend
Contesto Prompt
Injects a custom CONTEXT block into the Vision-Language prompt to steer the model towards the specific photographic domain of the archive. Without an active preset the behaviour is identical to the standard prompt.
Built-in presets: Nature Archive, Landscape & Territory, Astrophotography, Scientific Macro, Underwater, Reportage, Commercial, Street & Urban. The local LLM model can generate custom presets directly from the interface.
The active preset can also be overridden per-run from the gallery context menu, without changing the global preset.
Free pluginDownload and start.
Download OffGallery and bring your photo cataloging in-house.