A high-performance desktop asset manager built for the AI image generation ecosystem. It unifies metadata parsing across fragmented formats, providing SQL-backed search, Smart Collections, live folder monitoring, and AI-powered interrogation in a modern, multi-themed desktop interface.
Unified grid gallery with instant SQLite search and dynamic metadata parsing.
Virtualized grid browsing with AI-tag overlays, model/sampler/LoRA filters, and star ratings.
Duplicate Detective: Identify and manage identical or similar generations across your entire library.
Image Comparator: Pixel-peep fine details between two generations with the draggable slider.
View More Features (Collections, Speed Sorter, Scrubber & Settings)
Smart Collections: Create dynamic, auto-populating folders based on complex metadata filters.
Speed Sorter: Rapidly categorize massive generation dumps using keyboard hotkeys.
Metadata Scrubber: Strip hidden EXIF, prompts, and workflow data before sharing an image.
Settings: Control scanning behavior, manage the local database, and adjust appearance.
Designed for the privacy-conscious artist, this application operates on a strictly "Local-First" philosophy.
- Standalone Desktop App: Runs as a single
.exe(Windows),.AppImage(Linux), or.dmg(macOS). No installer required. - Bundled Runtime: Includes a self-contained Java 21 environment. No system-wide Java installation is required.
- Portable Data: All data (database, thumbnails, settings) is stored in a local
data/folder next to the executable, including on macOS (it sits beside the app bundle, not in~/Library/Application Support), making it easy to backup or move. - 100% Offline / No Telemetry: There are no "cloud sync" features, analytics, or background API calls. Your prompts and generation data never leave your machine.
- Privacy Scrubbing: Integrated Scrubber View allows you to sanitize images before sharing. It strips hidden generation metadata (Prompts, ComfyUI Workflows, Seed data) while preserving visual quality.
- Universal Metadata Engine: Advanced parsing strategies for the entire stable diffusion ecosystem.
- ComfyUI: Traverses complex node graphs (recursive inputs) and API formats to identify the true Sampler, Scheduler, and LoRAs used.
- Automatic1111 / Forge: Parses standard "Steps: XX, Sampler: XX" text blocks.
- Others: Native support for InvokeAI, SwarmUI, and NovelAI.
- Note: Metadata extraction requires images to contain embedded EXIF or PNG text chunks (standard for most AI generators).
- AI Auto-Tagger: Integrated WD14 ONNX model for local image interrogation. Automatically generate descriptive tags for your library without external API calls.
- Library Management:
- Smart Collections: Create dynamic collections based on metadata filters (e.g., "All images using Flux model with > 4 stars").
- Visual Previews: Collections feature a 3D-stacked image preview for immediate visual context.
- Pinned Folders: Bookmark frequently accessed directories for rapid navigation.
- Star Ratings: Rate images (1-5 stars) with instant filtering.
- Speed Sorting: A dedicated mode for processing high-volume generation batches.
- Hotkeys: Instantly move images to configurable target folders using numeric keys (1-5).
- Recycle Bin: Safely move unwanted results to the OS trash (Recycle Bin/Trash).
- Performance:
- FTS5 Search: Powered by SQLite's Full-Text Search for near-instant results across tens of thousands of images.
- Virtualization: Uses virtual scrolling to handle massive folders without UI lag.
- Project Loom: Uses Java 21 Virtual Threads for non-blocking background indexing.
- Modern UX & Customization:
- Latent Design System: A unified dark interface (cyan & violet accents) built from a shared component library for visual consistency across every tool view.
- Image Comparator: Side-by-side comparison tool with a draggable slider.
- Real-Time Hot Folder Sync: Keep the app open alongside your generator. A background file watcher detects new images and injects them into your gallery in real-time as they are generated, without locking the database.
- Network Drive & WSL Path Support: Folder monitoring works with any path your OS can browse to, including mapped network drives and Windows Subsystem for Linux shares (
\\wsl$\). Handy for pinning Dockerized ComfyUI/Automatic1111 output folders.
- OS: Windows 10/11 (64-bit), Linux (AppImage), or macOS (11+).
- Memory:
- Minimum: 4GB RAM.
- Recommended: 8GB+ RAM (especially when using the AI Auto-Tagger).
- Storage: ~300MB for the application + additional space for the WD14 AI model (~300MB) and thumbnail cache (~50MB per 10,000 images).
- GPU: Not required. AI interrogation runs efficiently on the CPU via ONNX Runtime.
The application is built as a highly concurrent hybrid desktop app, combining a resilient Spring Boot backend with a strict lazy-loading Vue.js frontend, packaged via Electron.
-
Backend (Java 21 + Spring Boot 3.3):
- SQLite + FTS5: High-performance local indexing and relational storage.
- Virtual Threads: Optimized for heavy I/O tasks (file scanning and metadata extraction).
- ONNX Runtime: Local execution of AI models with automated native resource management and idle-eviction.
- Flyway: Automated database schema migrations.
-
Frontend (Vue 3 + PrimeVue):
- Pinia: Centralized state management for the image library and UI state.
- PrimeVue: Premium UI component library with custom glassmorphism overrides.
- Vite: Modern build pipeline for the frontend assets.
-
Desktop (Electron):
- Process Management: Automatically spawns and terminates the Spring Boot backend.
- Native Integration: Provides access to native folder selection dialogs and OS file explorer.
- Cross-Platform: Builds for Windows, Linux, and macOS using GitHub Actions.
- Download the appropriate file for your OS:
- Windows:
Latent Library X.X.X.exe(portable, no installer) - Linux:
Latent Library-X.X.X.AppImage(mark as executable withchmod +x) - macOS:
Latent Library-X.X.X.dmg
- Windows:
- Run the application. No installation is required.
- Select a Folder containing your AI-generated images to start indexing.
🍎 macOS Users: Because this app is not yet signed with an Apple Developer Certificate, you may see an error saying the app is "damaged and can't be opened." This is a standard macOS security message for unsigned apps.
To fix this:
- Move the app to your Applications folder.
- Open Terminal.
- Run the following command to clear the quarantine attribute:
sudo xattr -cr "/Applications/Latent Library.app"- You can now open the app normally.
Latent Library is completely portable. To update to a new version without losing your collections, tags, or database:
- Download the latest release for your OS.
- Replace your old executable with the new one.
- Keep your
data/folder intact. As long as the new executable is placed next to your existingdata/folder, it will load your existing library automatically.
Distributed under the MIT License with the Commons Clause. Free to use, modify, and share. The Commons Clause restricts selling the Software or offering it as a paid hosted/consulting service. See LICENSE for the full text.
If Latent Library has been useful to you, consider supporting its ongoing development.
Developed by
Copyright (c) 2026 Alexander Nilsson