Lumina

Appearance

Technical FAQ ​

Where does Lumina store its models? ​

The models directory is resolved in this order:

  1. The LUMINA_MODEL_DIR environment variable.
  2. The location saved in Settings → Models.
  3. The platform default userData/models — %APPDATA%\Lumina\models on Windows.

Each model lives in a subfolder named after its id (for example anglenet/). In a source checkout, set LUMINA_MODEL_DIR in .env to keep using the repo's models/ folder.

What is an execution provider? ​

Models run as ONNX graphs through ONNX Runtime. The "execution provider" is the backend that executes the graph:

  • CUDA — NVIDIA GPUs (fastest on supported hardware)
  • DirectML — any DirectX 12 GPU
  • CPU — universal fallback

You can choose the provider per model. The installer you use determines which providers are available: the CUDA build ships CUDA support; the DirectML build ships DirectML + CPU.

Does Lumina run a server? ​

Yes, internally. Lumina spawns a small local FastAPI backend (localhost:8765) that performs all model inference and translation requests. It is started and stopped automatically by the app — you don't manage it, and it is not exposed to the network.

What is inside a .lmi project file? ​

A .lmi file is a zip archive containing:

  • project.json — manifest with page list, detections, layers, typography, and non-secret translation settings
  • copies of the source page images
  • inpaint mask patches (PNG)

API keys are never stored in .lmi files. See Projects & export.

How are translation API keys protected? ​

Keys are stored in Lumina's secrets vault and encrypted with the operating system's credential facility (DPAPI on Windows, Keychain on macOS, libsecret on Linux), persisted to secrets.json in the app's user-data folder. If the OS facility is unavailable, a base64 obfuscation fallback is used. Keys are only sent to the translation provider you configured, never to Lumina's servers (there are none).

Which models does Lumina support? ​

See the table in Managing models, including status, execution providers, and licensing notes.

Can I use a local LLM for translation? ​

Yes. The Custom provider works with any OpenAI- or Anthropic-compatible endpoint — including local servers like Ollama, LM Studio, or vLLM — by pointing Lumina at your own base URL. No text then leaves your machine.

How do I run Lumina from a source checkout? ​

See the Development section of the README: set up a Python 3.13 virtual environment, install the backend requirements, then build and start the Electron app. A GPU execution provider is optional for development.