TAU-HOME.COM
LOADING

Jasna: Local GPU-Accelerated Video Mosaic Restoration and Tool Architecture

An editorial analysis of Jasna, an open-source local GPU tool for targeted mosaic restoration in videos and images, covering hardware constraints and supporter

tau · October 19, 2026

#AI #Video Restoration #Open Source #BasicVSR++ #GPU

Jasna: Local GPU-Accelerated Video Mosaic Restoration and Tool Architecture

Jasna: Local GPU-Accelerated Video Mosaic Restoration and Tool Architecture

Image source: https://github.com/Kruk2/jasna

Jasna is an open-source, local GPU video restoration tool that detects mosaic regions and reconstructs missing details while keeping unaffected areas of the original picture untouched. It isolates target mosaic segments, providing high-fidelity local reconstruction through dedicated machine learning models without altering surrounding areas.

Targeted Video Restoration and Playback Workflow

Jasna identifies mosaic sections in videos and restores detail without touching unaffected frame areas. Users can isolate regions with the Segment Editor, which offers real-time previews before processing. The built-in Video Player allows live restoration playback without exporting to a file, eliminating unnecessary rendering cycles. Jasna includes BasicVSR++, a fast and free restoration model. Supporter tiers unlock LTX, a slower model with enhanced detail on NVIDIA GPUs. The tool also supports VR180 videos and provides an experimental SD 1.5 model for still images.

Hardware Prerequisites and Initialization Steps

Jasna requires specific local hardware and path configurations. NVIDIA users need a GTX 16-series, RTX 20-series, or newer GPU (compute capability 7.5+) with driver 610+ on Windows or 580+ on Linux; GTX 10-series cards are unsupported. AMD support is experimental for ROCm GPUs with Windows driver 26.8.1+, though secondary restoration and LTX remain NVIDIA-exclusive. The app folder must reside on an SSD formatted in NTFS, ext4, Btrfs, XFS, or ZFS without non-English characters in the path. On NVIDIA, the initial run spends 15 to 60 minutes preparing models for the GPU.

Supporter Models and License Allocation

Licensed under GNU AGPL v3.0, Jasna is free, using contributions to fund training GPU rentals. A 15 USD tier unlocks UNet 4x and SD 1.5, while 25 USD adds the LTX model and future updates. Backers who gave 15 USD or more on or before October 14, 2026 receive LTX for free. Genuine codes come exclusively from Buy Me a Coffee and Unifans; third-party sellers are fraudulent. A code activates up to three PCs or cloud workspaces and functions offline. Windows motherboard changes or Linux reinstallations consume a slot. Lowering Max Clip Size resolves VRAM shortages.

Sources

Impact-Site-Verification: 9b3ebd10-6c45-44cf-a014-5b8c19b7b285