Restore mosaic-covered areas in video and images using a local GPU tool
If you have masked specific parts of a video with a mosaic, you can locate those areas and restore them to their original state. Jasna is a free tool that uses a GPU to precisely restore only the mosaic regions in videos and images. It supports both an app and a command line, and runs directly on your computer.
How it restores only the mosaic
Jasna does not regenerate the entire video; it finds and restores only the parts processed with a mosaic, leaving the rest of the quality unchanged. The default model, BasicVSR++, is provided for free, while the supporter model, LTX, preserves more detail but processes more slowly and supports only NVIDIA GPUs. You can use the Segment Editor to preview and directly select the segments to restore. It can also handle VR180 videos and still images (SD 1.5, for supporters), and the Video Player allows you to view the restored video in real time without creating a file.
Runtime environment and initial setup
You need an NVIDIA GTX 16 series or RTX 20 series or higher GPU, with the latest drivers installed (Windows 610+, Linux 580+). AMD Radeon RX 7000/9000 series are supported experimentally, but some features are exclusive to NVIDIA. The save folder path must not contain non-English characters such as Chinese, Japanese, or diacritics, and network disks are not supported. SSD usage is recommended, as model loading is slow on hard disk drives (HDD). To use the LTX model, you need an NVIDIA RTX GPU, at least 8GB of VRAM, and at least 16GB of RAM.
What to know for the first run
Unzip the downloaded release file and run jasna.exe on Windows or the jasna file on Linux with NVIDIA. AMD Linux users should run run_jasna_amd.sh. The first run may be slow. For NVIDIA GPUs, preparing the model for the specific GPU takes about 15 to 60 minutes, and this process is performed only once. It is best to close other apps and not use the PC during this time. When updating, copying the model_weights folder to the new version can reduce this waiting time. If VRAM is insufficient, you must adjust settings such as lowering the Max Clip Size.
Supporter models and licensing
Jasna is free, but using additional models requires a license code. Donating $15 or more via Buy Me a Coffee, Unifans, or a China-region seller unlocks the UNet 4x and SD 1.5 models, while donating $25 or more unlocks the LTX model. Users who donate $15 or more before October 14, 2026, can receive LTX for free. A license code can be used on three PCs or cloud workspaces and works offline after activation. Be cautious of sellers outside official channels, as they may be scams.
Performance comparison and suitability
Comparing the restoration time for a 3-minute 22-second video on an RTX 5090 (Linux), Jasna v0.9.1 is 3.4 times faster than Lada 0.11.0 at 720p, 3.6 times faster at 1080p, and 4.7 times faster at 2160p. Lada uses less GPU memory than Jasna, but Jasna v0.9.1 has significantly improved memory efficiency compared to previous versions. If you need to remove mosaics from videos, it is suitable for users with NVIDIA GPUs. AMD GPU users should be aware of the experimental support, and those wanting the LTX model must verify they have an NVIDIA RTX GPU and sufficient VRAM. It is advisable to check in advance that you meet the path character restrictions and disk format (NTFS, ext4, etc.) requirements.
