Troubleshooting

If you run into a problem that isn't covered here, search the open issues on GitHub first, then open a new one with as much detail as you can — log output, your GPU, and which installer (DirectML or CUDA) you used.

Model downloads

Download is slow or fails. Models can be large (the vision-language OCR model is ~1.2 GB). Check your connection, free disk space, and retry from Settings → Models. Downloads can be resumed by retrying.

A model never shows as ready after downloading. Verify the models directory location under Settings → Models and confirm the files landed in the per-model subfolder (e.g. %APPDATA%\Lumina\models). You can also place model files manually there.

Performance and GPU

Everything is slow. Lumina falls back to CPU when no supported GPU is available, and some models are CPU-only by design (see the models table). Check that the execution provider is set per model in the settings, and consider the CUDA installer if you have an NVIDIA card.

CUDA errors or "provider not available". Make sure you installed the CUDA build (not the DirectML build) and that your NVIDIA driver is reasonably recent. If in doubt, the DirectML installer is the universal fallback and will run on any DirectX 12 GPU.

Out of memory during inpainting or OCR. Some models are memory-hungry. Try switching that model to a lighter alternative (for example lama_manga instead of paddleocr_vl), or close other applications while running heavy stages.

Translation

Translation fails with an auth error. Re-enter the API key in Settings → Translation. Keys are stored encrypted; if you reinstalled or changed user profiles the vault entry may be gone.

Custom provider returns errors. Check the base URL, model id, and that the endpoint is reachable from your machine. For Anthropic-compatible endpoints, enable the "Anthropic style" switch.

Translation is empty or obviously wrong. Try another provider/model, or review the instruction/style field. Translating with context (preceding bubbles) helps consistency — if the source language isn't detected well, set it explicitly.

The app or backend fails to start

• Make sure you installed with the matching installer for your GPU setup and that your antivirus/firewall isn't blocking the bundled backend or its local server.
• Fully quit and relaunch Lumina (including the tray icon if present).
• If the problem persists after a reinstall, open an issue on GitHub and attach the app log output.

Diagnosing issues

The fastest way to get help is to reproduce the problem and describe it precisely: which stage fails, which models are installed, your OS and GPU, and the exact text of any error message.