Turn 360p anime into 4K with Vulkan and a prayer

What it does Video2X upscales and interpolates video using ML models—think turning grainy 360p into sharper 4K, or 30fps into 120fps. It wraps several specialist engines (Real-ESRGAN, Real-CUGAN, RIFE, Anime4K) behind a single pipeline, with a Qt GUI for Windows and an AppImage for Linux.

The interesting bit The 6.0 rewrite is deliberately lean: it streams frames through ncnn + Vulkan without spilling intermediate files to disk, which matters when you’re processing hours of footage. The author notes version 5 “genuinely didn’t work”—this one claims to, and the hardware floor is surprisingly low (GTX 600-series, Intel HD 4000).

Key highlights

• Supports four distinct model families: Real-ESRGAN, Real-CUGAN, RIFE interpolation, and Anime4K v4 plus custom GLSL shaders
• Runs on Windows and Linux; Docker images for macOS/Linux, plus a Google Colab notebook for GPU-poor users
• Zero scratch-disk bloat during processing—final output only
• Precompiled binaries need AVX2 (Haswell/Excavator era) and Vulkan-capable GPU
• AGPLv3 licensed, with dependencies clearly tabulated

Caveats

• The demo videos are explicitly marked “outdated” (pre-6.0, using waifu2x and DAIN)
• The README’s “blazing fast” claim is unbenchmarked against prior versions or competitors
• macOS support is container-only; no native build mentioned

Verdict Worth a look if you regularly clean up archival or streaming-ripped video and want one tool that speaks multiple ML dialects. Skip if you need real-time processing or professional color-science controls—this is offline batch work, not a DaVinci Resolve plugin.