Discord: https://discord.gg/YHM7nFheyM
Additions:
- TensorRT support (8.6.0)
- Rife TensorRT (1080p max recommended input)
- Upscaling TensorRT (Custom models + Esrgan, some custom models might not work!)
- Adds option to disable frame extraction in rife-ncnn
- SPAN-ncnn
- Bugs (confirmed)
Changes:
- Swapped CX_Freeze to PyInstaller for CUDA build, please report issues that come up.
- Performance improvements across the board.
- Auto detect half precision mode, so supported GPUs will have it enabled by default. (Broken)
- Disable half precision button if your gpu does not support it. (Broken)
Fixes:
- Fix GMFSS (Broken)
Known Issues:
- CUPY does not recognize cuda install, so GMFSS is still broken.
- Auto detect half precision does not work.
- Inference with certain file names can stop scene change detection from working.
- Some distros have issues with TensorRT inference, please revert back to CUDA if it crashes.
Credits:
- https://github.com/NevermindNilas/TheAnimeScripter/ (For TensorRT implementation and some optimizations)
- https://github.com/adegerard (For many TensorRT optimizaitons)
Download Guide:
- For NVIDIA users, download the CUDA tar.gz file. Extract it, Note: Make sure all the files are together, THIS IS NOT AN APPIMAGE.
- Support for Ubuntu 22.04+ on CUDA edition because of packaging issues, NCNN version still supports 20.04+
- For MacOS users, refer to here
- For AMD and Intel GPU users, download the NCNN AppImage. Nothing has changed, fuse2 is required to launch the AppImage.