LuminaGetting started
1. Choose an installer
Lumina ships two Windows installers — pick the one that fits your hardware:
- DirectML installer — works on any DirectX 12 GPU and falls back to CPU. The universal choice.
- CUDA installer — NVIDIA-only, but noticeably faster on RTX/GTX cards.
DirectML runs on any DirectX 12 GPU and falls back to CPU. CUDA is NVIDIA-only but noticeably faster on RTX/GTX cards. Missing an asset? View the full release on GitHub.
Requirements
- 64-bit Windows (10 or newer is recommended)
- A GPU is optional — Lumina runs on CPU alone
- CUDA builds need an NVIDIA GPU with a reasonably recent driver
2. Install and launch
Run the installer and follow the steps, then launch Lumina. The app starts a small local backend (FastAPI + ONNX Runtime) automatically — no separate server or Python installation is required.
3. Download models
Models are not bundled with the installer. Open Settings → Models and download the models you need. The app groups them by stage:
| Stage | Recommended default | Notes |
|---|---|---|
| Text detection | rtdetr | Apache-2.0 licensed; rfdetr_seg adds segmentation masks but is restricted to academic, non-commercial use |
| OCR | baberu (multilingual) or manga-ocr (Japanese) | Choose based on your source language |
| Inpainting | lama_manga | Manga fine-tune; lama is a CPU-only general fallback |
| Aux | anglenet | Tiny (~2 MB) model downloaded automatically; measures text slant for typesetting |
Model files can also be placed manually in your models directory — see Managing models.
4. Translate a page
- Click Import (or drag images in) to add your first page.
- Open the page and run the stages in order — detection → OCR → translation → inpainting → typesetting. Each stage is a button in the pipeline panel.
- Translation needs a configured provider and API key — see Translation settings before you reach that step.
- When you're happy with the result, export the page as PNG or JPG from the Export dialog, or save the whole project as a
.lmifile.
For a detailed explanation of each stage and how to edit its results, read the Translation workflow guide.