Lumora

Local-first AI image gallery for your generated artwork. Fast, private, free.

What it does

Organize images from Stable Diffusion, Midjourney, ComfyUI, and similar tools — everything stays on your machine.

• Semantic search — find images by describing them, on a local Ollama or any OpenAI-compatible backend
• Visual search — CLIP-based similarity: "more like this", no text needed
• AI analysis — auto-tag, describe, and score (ratings stay human-only)
• Organize — smart collections, tags, favorites, ratings, trash with restore
• Export — custom filename templates, batch convert (jpg / webp / avif)
• AI agents welcome — built-in MCP endpoint for browsing, searching, and managing tags / favorites / trash
• Auto-updates — minisign-signed GitHub Releases

Quick start

npm ci
npm run dev

For local AI features, install Ollama and pull the models:

ollama pull nomic-embed-text
ollama pull llava

Prefer an OpenAI-compatible API (OpenAI, DeepSeek, Azure, local vLLM / llama.cpp)? Configure it in Settings → AI backend — the embedding and vision providers switch independently.

Keyboard shortcuts

⌘K opens the command palette; arrow keys navigate the gallery and sidebar; every operation has a shortcut.

Testing

npx vitest run              # 775 frontend tests
cd src-tauri && cargo test  # 272 Rust tests (269 passed, 3 ignored — need local Ollama)
npx tsc --noEmit            # type check

Documentation

• 📘 快速上手教程 — 10 分钟从安装到会用
• 📖 使用指南 — 完整功能参考（搜索/AI 分析/智能收藏/导出模板）
• 🏗 Architecture — 系统架构与数据模型

Architecture

Tauri 2 app: React frontend, Rust backend, SQLite database.

src/           React + TypeScript + Zustand
src-tauri/     Rust + SQLite + FTS5 + sqlite-vec

See ARCHITECTURE.md for details.

AI access (MCP)

The app exposes an MCP endpoint at http://127.0.0.1:{port}/mcp so AI agents can browse, search and organize (tags/favorites/trash) your library. Rating/scoring stays human-only. See docs/04-deploy/mcp.md.

License

MIT