Credits & Acknowledgements
OffGallery is made possible by extraordinary open source projects and public academic research. This page lists all third-party components with their authors, licenses, and official links.
Multilingual vision-language model for semantic search: converts images and text into comparable vectors in the same representation space. Trained on WebLI (10B pairs) with sigmoid loss — more accurate than CLIP on natural language queries and natively multilingual.
huggingface.co/google/siglip-so400m-patch14-384Vision foundation model for self-supervised visual similarity. Captures composition, texture and shape without semantic supervision. Used for visually similar image search.
github.com/facebookresearch/dinov2CLIP model specialized for biological recognition across 450,000+ Tree of Life species. Produces a full taxonomic hierarchy (kingdom, phylum, class, order, family, genus, species).
github.com/Imageomics/bioclipCLIP-based aesthetic scoring model trained on the LAION Aesthetics dataset. Assigns a 0–10 score for artistic and compositional quality.
github.com/christophschuhmann/improved-aesthetic-predictorMulti-Scale Image Quality Transformer: evaluates the technical quality of a photo (sharpness, noise, exposure) producing a continuous score. Implemented via the pyiqa library.
google-research / musiqPython library for Image Quality Assessment (IQA). Provides the interface for the MUSIQ model and other quality assessment models, with GPU support.
chaofengc / IQA-PyTorchOllama is the local LLM runtime. The Qwen-VL (Vision-Language) model generates descriptive tags, titles and textual descriptions by analysing photo content, entirely offline.
ollama.comAlternative local runtime for LLM models with an integrated graphical interface. Supported by OffGallery as a Vision-Language backend alternative to Ollama for generating tags, titles and descriptions.
lmstudio.aiDeep learning framework that powers all neural model inference (CLIP, DINOv2, Aesthetic, BioCLIP). Supports GPU acceleration via CUDA.
pytorch.orgLibrary providing access to pre-trained models (SigLIP so400m, DINOv2). Includes tokenizers, processors and a standardised inference API.
huggingface.co/transformersOpen source reimplementation of CLIP supporting many architectures (including ViT-L/14). Used by BioCLIP as the architectural backbone for biological training.
github.com/mlfoundations/open_clipPython bindings for the Qt6 framework. Provides the entire OffGallery GUI: windows, panels, tabs, image gallery, dialogs and all interactive widgets.
riverbankcomputing.com/pyqtrawpy is the Python binding for LibRaw, the C++ library for decoding 1000+ RAW formats (Canon CR2/CR3, Nikon NEF, Sony ARW, etc.). Enables thumbnail and metadata extraction from RAW files.
github.com/letmaik/rawpyThe modernised Python Imaging Library. Used to open, resize, convert and optimise images for the AI pipeline (pre-processing before embedding generation).
python-pillow.orgTool for packaging Python applications into standalone executables (.exe on Windows, binaries on Linux/macOS) without requiring an installed Python interpreter. Used to generate the OffGallery installer executables.
pyinstaller.orgThe definitive tool for reading and writing photographic metadata (EXIF, IPTC, XMP, GPS) across 300+ formats. Used to extract all metadata and write AI tags to XMP sidecar files.
exiftool.orgPython library for offline reverse geocoding: converts GPS coordinates (lat/lon) to city, region and country names. Geographic data (140,000+ locations) comes from the GeoNames dataset, bundled in the library — no network request required.
github.com/thampiman/reverse-geocoderFully offline machine translation engine based on OpenNMT models. Used to translate search queries into the tag language for keyword matching (SigLIP is natively multilingual and requires no translation for semantic search).
github.com/argosopentech/argos-translatePython environment and package manager. Used by the OffGallery installer to create an isolated, reproducible environment with all dependencies, without conflicting with other installed Python software.
docs.conda.io/minicondaFoundational scientific computing libraries for Python. NumPy manages float32 embedding arrays; SciPy provides distance and clustering algorithms used in semantic search.
numpy.orgEmbedded, server-less relational database. All OffGallery cataloging (EXIF metadata, vector embeddings, tags, scores) is saved in a single local SQLite file, with no database server required.
sqlite.orgOffGallery writes AI metadata (tags, descriptions, biological and geographic hierarchies) to XMP sidecar files compatible with the Adobe XMP standard. Hierarchical tags use the lr:hierarchicalSubject property, the de facto standard in the Lightroom workflow.
OffGallery was born from a passion for landscape photography, software development and new technologies. Designed, developed and maintained as an independent open source project.
github.com/HEGOM61ita/OffGalleryDeveloped the LM Studio plugin for OffGallery, integrating LM Studio as a Vision-Language backend alternative to Ollama. The plugin extends support to AMD GPUs via DirectML and to all compatible GGUF vision models.
theblackbird.altervista.orgOffGallery was developed with the support of various AI assistants for code writing, architectural review, documentation generation and problem solving.
Main AI assistant for interactive development. Claude Code supported architecture design, Python/PyQt6 code writing and review, debugging, documentation generation and website creation.
claude.ai · anthropic.comUsed for research, brainstorming and architectural approach comparison in the early stages of the project. Useful for exploring alternative solutions and getting a second technical opinion.
chatgpt.comUsed for technical research and library evaluation, particularly for HuggingFace ecosystem integration and complex data format analysis.
gemini.google.comUsed as an AI search engine to quickly find technical documentation, academic papers, library comparisons and answers to specific questions during development.
perplexity.aiStock photograph used in the banner at the top of the home page. The person depicted has no connection with OffGallery and does not endorse it.
PexelsStock photograph used, blurred, as the background of the banner at the top of the home page. Any brand visible on the camera body belongs to its owner and implies no relationship with OffGallery.
PexelsOffGallery is distributed under the AGPLv3 license. Third-party components listed on this page retain their original licenses, each indicated on their respective cards. The full text of each license can be found in the official repository of each project or in the
THIRD_PARTY.md file included in the OffGallery source code.Registered trademarks mentioned (OpenAI, Meta, Google, Adobe, Hugging Face, Anaconda, etc.) are the property of their respective owners and are referenced solely for informational/attribution purposes.