Lumina v0.2.1

⚠️ 0.2.0 withdrawn — the 0.2.0 installers were pulled due to installer bugs (oversized app.asar and a console window popping up during CUDA extraction). The 0.2.0 changelog entry remains for reference at CHANGELOG.md#020---05-09-2026.

Fixed

• CUDA installer no longer pops a console window: extracting the ~1.5 GB CUDA runtime used to open a separate terminal during setup. Extraction now runs inside the installer process itself, driving the native progress bar — no terminal.
• Installers are dramatically smaller: the app package (app.asar) was accidentally including the previous build's entire unpacked output (a second full Electron runtime, ~1.7 GB). It's now limited to the actual app code (~2 MB), shrinking both the DML and CUDA installers significantly.

Lumina v0.1.0-experimental-preview

Lumina 0.1.0-experimental-preview

Initial experimental preview release.

Installer

Windows installer only: Lumina-Setup-DML-0.1.0-experimental-preview.exe

Built and published automatically via GitHub Actions on every version bump.

GPU acceleration

This installer bundles the DirectML execution provider only.

• Runs on any GPU that supports DirectX 12, including NVIDIA
• Falls back to CPU automatically on machines without a compatible GPU
• CUDA acceleration is not included in this installer. It is only available when running from source.

Some models have fixed provider behavior regardless of the selected runtime: lama runs on CPU only, and lama_manga / paddleocr_vl never use DirectML.