Changes
vs_temporalfix is now on PyPI! Which means it can be installed with pip install -U vs_temporalfix and (almost) all dependencies will be automatically installed too. Check the readme for details. This requires Vapoursynth R74 or newer.
This update also brings lots of other changes:
Temporalfix AI Model
- Strength now goes smoothly from
0.0-3.0, which means it can be precisely tuned with values like1.5and goes down to0. This adds a new requirement:pip install onnx - Added support for TensorRT 11 and the
tensorrtpython package for engine building. Trtexec is also still supported. - Fixed
excludesometimes not working with some input formats when using the CUDA backend. - Better error messages.
Temporalfix Classic
- This update changes the output slightly. It didn't cause issues on my test clips, but please report any regressions.
- About 25% performance improvements and RAM usage reduction mainly by doing prefilter steps at lower chroma resolution.
- Better handling of subsampled and gray inputs, resulting in total performance improvements of 75% for 4:2:0 subsampling and 150% for gray (that's 2.5 times as fast). A bit less in 4k.
- Fixed error when the input is in a 16-bit float format and
debug=True. - Replaced no longer maintained plugin tcanny with edgemasks.
- Replaced no longer maintained plugin motionmask with akarin expr.
- Replaced no longer maintained plugin retinex with akarin expr.
- Plugin vszip is now mandatory and no longer optional.