jasna: A tool that finds and restores only mosaic areas in videos on GPU
Kruk2/jasna
About the project
Jasna is a tool that automatically finds and restores mosaic areas in videos and images. It preserves the quality of the non-mosaic parts exactly as in the original and handles the entire process from video decoding to encoding on the GPU. It requires an NVIDIA GTX 16 series or higher, or an AMD Radeon RX 7000/9000 series GPU, and must be installed on a local SSD storage.
The default model, BasicVSR++, is provided for free and boasts fast processing speeds. If you want higher detail, you can use the supporter-exclusive LTX model. This model is diffusion-model-based, so processing time is longer, but it offers excellent detail description. You can directly specify the segments to restore using the Segment Editor, check them via preview, and then execute the task.
Beyond simple restoration, you can integrate auxiliary restoration models such as UNet 4x or Topaz TVAI to increase resolution and improve sharpness. It also supports VR180 videos and still images (SD 1.5 model), and you can view real-time restoration results immediately without generating files through the Video Player. Batch processing is also possible via the command-line interface.
It is suitable for developers who want to perform video restoration while maintaining privacy in a personal GPU environment. Initial model preparation for the GPU takes 15 to 60 minutes, but afterwards, you can copy the model_weights folder to reuse it, reducing wait times. It is distributed under the AGPL-3.0 license, and a separate license code is required to use the supporter model.
Kruk2/jasna
Video restoration tool
Python