Downscaling outrageously slow

I just upgraded to version 3.1.0 and I’m finding that using upscaling is very very slow. By “upscaling” I’m converting a 1920x1080 video to HD (1280x720). I’m not changing the FPS and letting TVAI chose the Proteus fine-tune/enhance model.
Does anyone have any ideas? I just paid $149 to upgrade my license so I could run this release and so far it seems like a step backwards unless I’m doing something really stupid.

Downscaling, or did you mean 1280x720 to FHD 1920x1080?

In past versions, there is a model selecting bug where the 4k version of the model was basically always selected. I have not tested with 3.1.0 to see if it’s still there, but it was there in the last version 3.0.12.

This post, sounds very similar to what’s going for you. If I’m not mistaken, it’s upscaling to 4k then downscaling to the output resolution using a common lanczos resizing filter.

I did mean downscaling, sorry for the confusion. 1920x1080 → 1280x720. It’s processing 0.2fps; a one minute and two seconds MP4 file is scheduled to take 3 hours. I’m pretty sure the version I just upgraded from (3.0.12) didn’t exhibit this behavior.

As a sanity check, I uninstalled 3.1.0, reinstalled 3.0.12, and now the same video that would have taken 3 hours to downscale is scheduled to take 48 minutes. I’m hoping someone can test this for me to confirm.

Maybe it’s just the downloaded models that are faster.

That’s possible but how does that help me with the new version? To add salt to the wound, at 11:30AM today I purchased the upgrade license and at 3:30PM I got an e-mail from Topaz telling me I could upgrade for $119…and I’m not even using the new release now. It hasn’t been my day.

BTW: Topaz did refund me the $30.

Why on earth would you use TVAI for downscaling?! I doubt the models are even trained for that; but, more importantly, you can do so, blistering fast, with any AVI/VapourSynth script filter.

I agree with that statement, but maybe they want some benefits of one of the AI models on the downscaled version. Maybe the real issue here is the GUI locks the enhancement model in place when any resolution change is selected. When downscaling, ffmpeg alone would do a super fast job.

1 Like

That’s a fair question; I use it to sometimes downscale and enhance (remove noise, etc.) from the video. Having said that, it ought to be able to downscale without issues.