nunif is a collection of image and video processing tools built around deep learning models for anime and photographic content.
The project addresses several distinct problems. waifu2x performs super-resolution upscaling specifically tuned for anime-style artwork and photographs using PyTorch implementations of GAN-based models. iw3 converts 2D images and videos into stereoscopic 3D format for viewing on VR devices, with variants including iw3-desktop for real-time streaming of PC screen content and iw3-player for self-hosted playback through WebXR. stilizer provides fast video stabilization using conservative algorithms. cliqa assigns image quality scores useful for filtering low-quality images during dataset creation, supporting JPEG quality prediction and noise level estimation.
Adoption depends on your specific use case. waifu2x suits anyone upscaling anime artwork or photographs who wants a PyTorch-based implementation with pretrained models; cloud and in-browser demo versions are available for testing. iw3 is designed for VR enthusiasts wanting to watch arbitrary 2D content as stereoscopic video on VR devices, particularly those with PC-based workflows. stilizer addresses video stabilization needs. cliqa helps with dataset curation by filtering images below quality thresholds. The project explicitly states it is a personal playground where incompatible changes may occur. Installation requires Python 3.10 or later and PyTorch, with platform-specific guides for Windows, Ubuntu, and macOS, plus optional support for Intel GPUs and older NVIDIA hardware. A Windows installer package is provided for end users.
Development is exploratory and subject to breaking changes. The maintainer describes the repository as a playground and explicitly warns that incompatible changes will be made. The codebase spans multiple distinct tools rather than a single focused project, suggesting experimentation across different problem domains. Documentation is distributed across separate README files for each component, and the project maintains flexibility around dependencies and build configurations through environment variables and optional installation paths.
