Topaz Video AI Linux Beta v5.0.3.0.b

Hello everyone,

We have a new beta available for our Linux users.

5.0.3.1.b
Ubuntu 22.04 | SHA-256

Changes from 5.0.3.0.b:

  • Fix models failing with CUDA Out of Memory when running on NVIDIA GPUs when TensorRT models are unavailable

5.0.3.0.b
Ubuntu 22.04 | SHA-256

Linux Specific Changes from 4.2.2.1.b:

  • More models should now be able to run when TensorRT versions of that model are not available

Changelog from 5.0.2

  • Added back reset button for timeline zoom.
  • Added .SER as an importable file type.
  • Fixed certain input tiff sequences failing to load with projects.
  • Fixed 0s elapsed bug when finishing export.
  • Fixed stabilization not reloading in app (loading with error).
  • Fixed benchmarking failing to load.
  • Fixed trim + crop + rotation persistence across projects.
  • Fixed EXR+DPX failing to reconvert.

Known Issues

  • EXR/DPX Exports may lose input frame number.
  • Inconsistent “Preview X frames” enable status.
  • Looping previews has a stutter.
  • Live preview can have frame de-sync.

Known Issues (Linux Specific)

  • 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
10 Likes

I can’t get this build to open the browser for activation. It doesn’t report any errors in the GUI or the console, just fails outright.

I had to check the log in ~/.local/share/Topaz\ Labs\ LLC/Topaz\ Video\ AI\ Beta/logs/ to find the URL, manually copy/paste that to the browser and then the activation worked.

This was true under Wayland also for QT_QPA_PLUGIN=xcb

Having activated it, using ‘browse’ to try and find a video file, the tool instantly crashes. The log shows:

qrc:/qt-project.org/imports/QtQuick/Dialogs/quickimpl/qml/FileDialog.qml:4 Cannot load library /opt/TopazVideoAIBETA_5.0.3.0.b/bin/Qt/labs/folderlistmodel/libqmlfolderlistmodelplugin.so: (/opt/TopazVideoAIBETA/lib/libQt6Core.so.6: version `Qt_6.7’ not found (required by /usr/lib/libQt6LabsFolderListModel.so.6))

This may be due to Arch/EndeavourOS. Drag-and-drop works fine.

On my 30 series card, the AI engine fails for both Proteus and Gaia, outright. Artemis seems to run on the GPU.

Logs from failed runs


2024-05-03-09-28-12-Main.tzlog (181.4 KB)

Are you using KDE? I recall that that also uses Qt, so that may be the cause of the UI issues you’re having there if it’s expecting a different Qt version than we use.

Thanks for the logs; we’re going to keep looking at what’s going on with the non-TensorRT models.

Do you have logs from attempting to run that model?

Arch/Endeavour releases Plasma built against Qt 6.7 from what I can see.

It won’t fix the conflict between Plasma’s Qt and ours, but could you try this alpha build and see if you’re able to run Proteus and/or Gaia now?

This build found the AMD graphics initially, so I switched back to the 3070 RTX in preferences. Both attempts, for multiple models, end up on the CPU only.

2024-05-03-18-17-53-Main.tzlog (743.1 KB)
2024-05-03-18-26-27-Main.tzlog (833.1 KB)

It seems like I’m unable to download this model specifically, this also happened with previous beta builds. I will try again soon and probably share logs after.

Able to activate successfully. Generally, things seem functional, but the scale of the application seems a little to large. The text is too big, buttons and boxes are too big, etc.

1 Like

Should I be able to run Proteus on GPU with this release?

Oops. Rather embarrassing typo under the hood on my end there. Can you try this build instead?

Yes, you should be able to. If you’re not having any luck, could you try the build I just linked Phil right above this message?

1 Like

That looks quite a bit more promising. Gaia is running at @1.7 fps on a 1920x1080 on my RTX 3070. Not sure how to tell whether it’s using TensorRT or CUDA, but assume CUDA because the framerate isn’t great.

Proteus runs at 3.5-3.7 fps on the same footage.

Curiously, Gaia preview seems to be temporally offset from the source; using the wipe comparison, the preview is coming from 1 or more frames later in the footage. The Proteus preview appears to be temporally consistent with the source footage.

Quick bug report. When I go into Preferences and change project or model directories, the text box after selecting the folder is missing the first /, making the paths invalid. If I manually add it, they work.

Just tried it, works amazing under
System:
Kernel: 5.15.0-106-generic x86_64 bits: 64 compiler: gcc v: 11.4.0 Desktop: GNOME 42.9
tk: GTK 3.24.33 wm: gnome-shell dm: GDM3 42.0, LightDM 1.30.0 Distro: Linux Mint 21.2 Victoria
base: Ubuntu 22.04 jammy

The only thing might be, that Linux limits it in disk access and in general access.

I can’t really say what it is, but already exported video parts don’t get updated, if i export them again with different resolution and same file name in same folder.
I do this so i also have audio, instead of preview, somehow i just need this.

Maybe Linux does not allow overwrite files, never had that problem in Windows.

Maybe i have to start in root, which i would obviously not like to do right now.

The missing permissions might also be indicated by the fact, if you open your Preferences and choose another folder for your workspace folder, right click on that folder choose “Copy Location” so you can copy paste that location into your file browser, it freezes.

Also seems to have to re download Models every time.

Just as a feedback, it’s amazing alone that i can use it under Linux now.

1 Like

Just got my new GeForce RTX 4070 card yesterday, and been testing today with the Linux beta. I’m getting an fps of 18 with Artemis on low quality HD video, so that’s reasonably good. It seems to be working smoothly from command line and GUI.

I was wondering if it would be possible to build the shipped ffmpeg with libx265 support? The NVIDIA H265 does not work with the drivers shipped with Debian 12.