Topaz Video AI Linux Beta v4.2.2.0.b

Hello everyone,

We have a new beta available for our Linux users.

4.2.2.1.b
Ubuntu 22.04
SHA-256: 564415a97b31723f421d99e85e97902a6f5521e8fe8acc726355b73f0f47f859

Changelog from 4.2.2.0.b to 4.2.2.1.b

  • Fixed interlace model resetting to Prob4
  • Fixed jittery output with Apollo Fast
  • Updated videoai + OpenVINO

4.2.2.0.b
Ubuntu 22.04

Changelog from 4.2.1

  • Add Upscaling to OpenFX plugin.
  • Fixed app hanging when swapping inputs quickly.
  • Fixed preset issue with deinterlace 2x being overwritten to 1x.
  • Fixed problem where batch processing will only export one video.
  • Updated minimum trim length from 1sec to 5 frames of time.
  • Fixed inconsistent model labels in enhancement and second enhancement.
  • Fixed issue with ProRes reporting two Codec labels.
  • Fixed bug where opening preferences windows would stack infinitely.
  • Removed incompatible settings with mkv container.

Linux Specific Changes

  • Includes Windows/Mac changes since last Linux beta
  • OpenFX plugin (beta)
  • In-app updater support

Known Issues

  • EXR/DPX Exports may lose input frame number.
  • Inconsistent ā€œPreview X framesā€ enable status.
  • Looping previews has a stutter.
  • Live preview can have frame de-sync.

Known Issues (Linux Specific)

  • Models that do not support TensorRT may not run on NVIDIA GPUs.
  • The OFX plugin is installed to /opt/TopazVideoAIBETA/plugins/TopazVideoAIBETA.ofx.bundle. You may need to symlink this into /usr/OFX/Plugins/ to use it in your host application
  • Davinci Resolve is not currently a supported host on Linux
  • Nyx 4x option in OFX plugin will not work
12 Likes

@gregory.maddra Any plans to offer other than Debian based Linux systems ?
I run Arch alongside windows.

Thank you for supporting Ubuntu. I look forward to support for the next LTS release and I would be happy to stay a customer for that reason.

Not at the moment. However, I’ve seen a few people in the previous beta threads mention they’d managed to get it running on Arch. I believe they used debtap to convert our package to an Arch package, but I can’t guarantee that won’t break anything.

I can confirm this works for the most recent beta. I did need to remove ā€œgediā€ as a package from the .PKGINFO file. I’m at work right now so I’m not 100% on the details but I believe that’s correct.

Indeed, this works for me on Arch. It needs LD_LIBRARY_PATH set, because Arch puts cuda outside the regular path (/opt/cuda) so I run with :

LD_LIBRARY_PATH=/opt/cuda/lib:/opt/cuda/bin:/opt/TopazVideoAIBETA/lib:/opt/TopazVideoAIBETA/bin /opt/TopazVideoAIBE
TA/bin/Topaz\ Video\ AI\ BETA

I don’t tend to use debtap at this point. I use Ark to extract the .xz archives from the .deb file and then just slot those files into the file system manually.

1 Like

I removed ā€œgediā€ as a dependency but it needs a lot more removed.
This was on CachyOS.

One thing that is tripping me up with this build is that attempting to login constantly fails:

/usr/bin/xdg-open: line 745: 850813 Aborted (core dumped) kde-open "$1"

Are you using Wayland by any chance? We recently started including the Wayland platform plugin rather than just X.

If so, does the login button work if you run after setting the environment variable QT_QPA_PLATFORM=xcb?

I am indeed. Providing that for the regular application did not seem to help - the activate and login buttons silently failed. So I then tried this:

LD_LIBRARY_PATH=/opt/cuda/lib:/opt/cuda/bin:/opt/TopazVideoAIBETA_4.2.2.0.b/bin:/opt/TopazVideoAIBETA_4.2.2.0.b/lib QT_QPA_PLATFORM=xcb TVAI_MODEL_DIR=/opt/TopazVideoAIBETA/models /opt/TopazVideoAIBETA_4.2.2.0.b/bin/login

This output:

LOGIN|https://topazlabs.com/oauth/authorize?client_id=XbU0xqXX0H1U4GsI3B3xIzJwVY3KL3grXhUTm28K&response_type=code&state=53699|
qt.qpa.plugin: Could not find the Qt platform plugin "xcb" in ""
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

/usr/bin/xdg-open: line 745: 1172415 Aborted                 (core dumped) kde-open "$1"

I could click on that link and allow login to go ahead in the browser. However, the main GUI doesn’t seem to see the activation status, so I get prompted to login/activate and this doesn’t work.

I wouldn’t entirely be surprised if this is some weirdness between whatever version of Qt KDE is using and the version we’re shipping. I’m afraid I don’t have a system using KDE to experiment with for this, so I’m guessing a bit here.

Could you try also setting QT_PLUGIN_PATH=/opt/TopazVideoAIBETA/bin? As long as you end up with an auth.tpz file writen in /opt/TopazVideoAIBETA/models the GUI should be able to pickup the login.

Specifying :

LD_LIBRARY_PATH=/opt/cuda/lib:/opt/cuda/bin:/opt/TopazVideoAIBETA_4.2.2.0.b/bin:/opt/TopazVideoAIBETA_4.2.2.0.b/lib QT_QP
A_PLATFORM=xcb QT_PLUGIN_PATH=/opt/TopazVideoAIBETA_4.2.2.0.b/bin /opt/TopazVideoAIBETA_4.2.2.0.b/bin/Topaz\ Video\ AI\ BETA

seemed to bring it up and it appeared to find the existing activation (no activate or login dialog was then shown)

1 Like

I need to qualify this. I can actually get the GUI up, but 4.2.2.1.b doesn’t actually seem able to load any model on to the card and run it. I get a ā€˜failed to autocorrect error’ dialog that doesn’t tell me what the error is. I’m simply trying to run Artemis with default settings on a 25 fps 1440x1080 video.

The previous Linux ā€˜beta’ 4.0.7.0.b comes up and processes this same video without any issue at all, with the exact same settings, in the exact same environment, on exactly the same hardware. This makes me sad.

In trying to figure out the failure for 4.2.2.0.b, I can’t get the logs up either. The GUI just reports an ā€œAI failureā€ and the ā€˜get logs’ button and related menu options do nothing:

LD_LIBRARY_PATH=/opt/cuda/bin/:/opt/cuda/lib/:/opt/TopazVideoAIBETA/bin/:/opt
/TopazVideoAIBETA/lib/ QT_PLUGIN_PATH=/opt/TopazVideoAIBETA/bin /opt/TopazVideoAIBETA/bin/Topaz\
Video\ AI\ BETA
qt.qpa.wayland: No shell integration named "xdg-shell" found
qt.qpa.wayland: No shell integration named "wl-shell" found
qt.qpa.wayland: No shell integration named "ivi-shell" found
qt.qpa.wayland: No shell integration named "qt-shell" found
qt.qpa.wayland: Loading shell integration failed.
qt.qpa.wayland: Attempted to load the following shells QList("xdg-shell", "wl-shell", "ivi-shell"
, "qt-shell")
qt.qpa.plugin: Could not load the Qt platform plugin "wayland" in "" even though it was found.
default font 9 "Sans Serif"
default font 10 "qrc:/elements/Roboto-Regular.ttf"
2024-03-29-9:04:19.715 default warning 3632-3632 unknown unknown:0 - QVariant::load: unable to l
oad type 39.
2024-03-29-9:04:19.715 default info 3632-3632 unknown unknown:0 - Deleting files "/home/phil/.lo
cal/share/Topaz Labs LLC/Topaz Video AI Beta/logs/2024-02-14-20-51-33-Main.tzlog"
[swscaler @ 0x7c8e0412e5c0] No accelerated colorspace conversion found from yuv420p to rgba64le.
qt.qpa.wayland: No shell integration named "xdg-shell" found
qt.qpa.wayland: No shell integration named "wl-shell" found
qt.qpa.wayland: No shell integration named "ivi-shell" found
qt.qpa.wayland: No shell integration named "qt-shell" found
qt.qpa.wayland: Loading shell integration failed.
qt.qpa.wayland: Attempted to load the following shells QList("xdg-shell", "wl-shell", "ivi-shell"
, "qt-shell")
qt.qpa.plugin: Could not load the Qt platform plugin "wayland" in "" even though it was found.

Log, in case useful

2024-03-29-09-04-19-Main.tzlog (104.2 KB)

It looks like there isn’t a TensorRT version of that model for that GPU on Linux yet. Running models that don’t support TensorRT on Linux is still somewhat iffy.

We have some changes planned to try and address this in a future release.

Just popping in to say that the model download error is intermittent and seems to be related to the actual API its calling to download the models. It’ll fail, and then when I do the same thing a few minutes later it will succeed. But not for all models - iris LQ will never download (nvidia 4090).

I spoke too soon, after a couple of hours of trying Iris LQ did come down.

This is a weird problem. Models that downloaded yesterday are trying to redownload and fail. Today’s struggle seems to be specific to Chronos Fast (linux / nvidia 4090). If I go into model manager and try to pull that model down, it fails every time with ā€œConnection Handling Canceledā€. But I was able to pull down other models, including (regular) Chronos

Same problem with the stabilization model as well.