Skip to content

Troubleshooting and FAQ

mafiosnik edited this page May 15, 2023 · 5 revisions

Why do I have to convert engines for CAIN (TensorRT) multiple times?

This is a workaround to make CAIN work on TensorRT inference. It doesn't support dynamic shapes, and you have to convert an engine file for every new video resolution. It's a bit inconvenient at first, but definitely worth the 4x-5x speed boost compared to NCNN, especially from the point on once all the engines for your most used resolutions are converted.

Where can I get custom upscaling models?

I can't include most models in enhancr because of Licensing. Custom models can be obtained here and loaded inside the models tab.

I'm running out of memory when trying to run a TensorRT inference, what should I do?

Either lower concurrent streams in settings or set up tiling in combination with a smaller engine resolution to decrease Video Memory usage.

I get "Python exception: operator (): no valid optimization profile found"

This means that shapes are not fitting, make sure you are using custom shapes in settings if you go over 1080p input or use correct tiling settings if you use tiling.

I get "[h264/h265/av1_nvenc @ xxxxxxxxxxxxxxx] No capable devices found"

If you are upscaling, take note that the NVENC encoder has a resolution limit of 4096x4096 for h264_nvenc and 8192x8192 for h265_nvenc/av1_nvenc. If the output exceeds this you will encounter this message.

Take into consideration to use software encoding instead of hardware encoding, due to encodes made on hardware looking worse by a lot and way larger file sizes over software encoding.

Furthermore this can happen when there are simply not enough NVENC Streams available, because other applications take them up, or when your GPU doesn't support NVENC. Enhancr encodes Preview on hardware by default now, so if you still encounter any issues even with disabled hardware encoding, please disable Preview in the settings.

"No attribute with the name lsmas exists" with SVP installed

Simply uninstall SVP, I currently have no idea what they are doing to cause this interference. Regardless, it has been proven multiple times in the past that this solves it, if you didn't mess with enhancr's files before.

Clone this wiki locally