Topaz Video AI 4.2.0.0.b - 4.2.0.1.b

Hello everyone,

We have a new beta available for testing with some significant new features.

4.2.0.1.b

Changelog from 4.2.0.0.b

  • Stability and performance improvements for Resolve plugin.
  • Slow motion dialog box accepts custom values again.

4.2.0.0.b

Changelog from 4.1.2

  • First beta plugin release for Davinci Resolve
    • Video AI is now available within Davinci Resolve as an OpenFX plugin. This first release includes all Enhancement models + Motion Deblur for both Windows and macOS.

    • The OpenFX plugin can be installed either during app installation (automatic on Windows and an optional checkbox on macOS) OR after app installation from the “Plugins” button in the app’s top menu bar.

    image

    • The OpenFX plugin can be used in both the Edit workspace and as a Node within the Color workspace

  • Aion Frame Interpolation model
    • More accurate motion estimation for high-resolution inputs.

    • Reduced artifacting and ghosting when compared with Apollo and Chronos.

    • True 16x slow motion without duplicate frames.

    • Improved temporal consistency.

    • Can reduce tiling artifacts depending on available VRAM.

    • No TensorRT support in this beta release.

  • 3D LUT option in Export panel
    • Process log footage directly in Video AI by selecting a .cube file to apply during exports

    image

  • Low Disk Space Warning System
    • When queuing items for export, Video AI will now warn if the selected output disk is low on space. This threshold is fully customizable from the Preferences → Advanced menu.

    image

Other Changes & Fixes

  • Changed Previews/Exports sections to use tab buttons to toggle display.
  • Fixed bug where renamed FHD presets still have the previous preset as a duplicate.
  • Fixed bug where some codec names caused the export button to overlap.
  • Fixed bug where file names containing '#' were truncated in "Export As" dialog.
  • Fixed bug where FPS is doubled when reading interlaced video files and caused errors for Frame Interpolation.
  • Added options to append original filename and model names to image sequence export settings.
  • Fixed various rotation issues.
    • Fixed bug where cropping a rotated image sequences did not display the full frame.
    • Fixed bug where Auto Letterbox/Pillarbox crop was done incorrectly for rotated videos.
    • Fixed bug where switching inputs would reset rotation in preview views.
    • Fixed bug where input video thumbnails did not correspond to changed rotation.
    • Fixed bug where rotation metadata was not read when files were imported.

Known Issues

Resolve-specific

  • On first launch, you may need to manually enable the Video AI OpenFX plugin from Resolve's preferences menu:

  • TensorRT is not currently supported on NVIDIA cards.
  • It is recommended to enable "Show All Video Frames" in the Edit workspace when using the Enhance plugin.
  • image

  • Nyx performance is significantly slower compared to the in-app model.
  • Video AI app

  • Batch processing selection can overwrite FPS output.
  • 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.
6 Likes

Very excited to try the resolve plugin. This could be amazing for my workflow

Please do not dim down the preview windows, I can barely see anything.

Sometimes the zoom level jumps in and out for a millisecond when moving over the end of the preview and back into the preview again.

I made the in app update. Everything is okay. My preferences are the same as before, but the export options changed from h264 mpg (before update) to h265 mkv (after update) unexpected.

2 Likes

The audio exports to ipcm format even though the audio is set to “copy”. VLC player can play it but MPC can’t.

The progress bar during export is also gone.


Maybe it’s time to put some of those interpolation models on another line?

2 Likes

Cannot enter 2.5 in the slow motion box anymore. I can put it there,
image
but it goes away once I click enter to make it apply.
image

1 Like

every time again I sort the benchmark data by hand. :neutral_face:

Topaz Video AI Beta  v4.2.0.0.b

System Information
OS: Windows v11.23
CPU: 13th Gen Intel(R) Core(TM) i9-13900K  31.775 GB
GPU: NVIDIA GeForce RTX 4090  23.576 GB

Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080

Benchmark Results
Artemis		1X: 	40.75 fps 	2X: 	18.35 fps 	4X: 	05.06 fps 	
Iris		1X: 	36.38 fps 	2X: 	23.25 fps 	4X: 	05.74 fps 	
Proteus		1X: 	38.91 fps 	2X: 	22.49 fps 	4X: 	05.92 fps 	
Gaia		1X: 	15.24 fps 	2X: 	10.45 fps 	4X: 	05.34 fps 	
Nyx		    1X: 	17.10 fps 	2X: 	13.70 fps 	
Nyx Fast	1X: 	30.05 fps 	

4X Slowmo		
Apollo: 	45.29 fps 	
APFast: 	76.39 fps 	
Chronos: 	32.22 fps 	
CHFast: 	35.12 fps 	
Aion: 	    17.02 fps 	

1 Like

I’m having system crashes: computer reboots when a render with aion finishes. Not always. About 70% of the renderings. In these cases, sometimes the output video file is fine, sometimes the output cannot be opened with a media player. I’m processing 4K videos, but I can’t find exactly under what circumstances the system crashes and under which it’s fine. But it occurs even if aion is the only model used.

Also sometimes if I stop the process, the reboot comes up.

In the logs the last rendering ended with a OS (Sonoma 14.3.1) restart.

Additional note: I needed to clear all coreMLCache in the models folder for aion to work. Maybe it should be enough to delete only the aion model subfolders (aion and aionb), but it didn’t work. If you have that problem (related to a previous aion installation), you can try deleting only both specific subfolders first.

logsForSupport.zip (100.1 KB)

@tony.topazlabs

I did a quick test with the new AION, using video downloaded from YT at 1080p 60fps resolution (if need be then I can even link it), and to convert this video of about 3 minutes length to 4x slowmotion, with my mid-range Windows 11 PC, the total conversion would take more than 2-3 hours, while with the APOLLO model in 4x slowmotion took about 20 minutes.

I would say that as performance is still not there…I also noticed that my RTX 3070 FE, coupled with a Ryzen 9 3900x Processor in full mode, was not being used properly (the usage percentage in HWInfo was about 79-81 percent)

There is still a lot of room for improvement…

1 Like

TensorRT is not currently supported on NVIDIA cards in this beta!

1 Like

From the way it is written in the changelog, I think this limitation only applies to the DaVinci Resolve Plugin… otherwise it should have been specified that it applied to the AION model!

@tony.topazlabs this major Bug was not fixed in this beta? why? basically you can not use Batch video process in interlace mode, it resets all the settings.

1 Like

Yeah, they’ve been needing to do it even before now.

That’s how it reads to me too.
I sure hope that the speed can be improved, but also the grainy blur is still very prominent. Fine for increasing frame rate. Terrible for slow motion.

I’ve just edited the post to clarify this point. TensorRT is currently not supported in the Resolve plugin AND it is not offered in the current beta version of Aion. We plan to ship a TensorRT version of Aion for the main release.

2 Likes

This is an area our team is currently working on, we expect to have batch processing issues resolved in a patch release between 4.2 and 4.3

2 Likes

First run on MacOS successful. Thanks.

Am I missing something or there is still no live preview during export? Also tried to open the video while processing (on Mac) and it won’t open. So “Known issues” should state “Live Preview Not Available”.

This problem is easily solved by using the mkv container instead of mov.

The very first thing that stuck out that is VERY easy to reproduce, putting in a manual playback zoom no longer works. It now has Fit, 100%, 200%, and 400%. I like to type in 50% a LOT as it is the perfect zoom to see a larger picture for any artifacts, then 100-200% for fine tuning. So that’s an issue.

#2 is looks like Aion V1, and every single other model, has HUGE issues on anything not people with lots of movement.

That is a pretty large nope there :wink:

Also, the playback is back to being all messed up. This time I am on my bic boi machine, Ryzen 7 5800x, 128gig 3600 ram, Nvidia 4070 with all nvmes. I counted about 5-15 seconds when I push play to when it plays. This also include trying to move around the playback slider to any point in the video. My slower rig (i7-6700k, 32gig ram, 3060) takes anywhere from 30s-1.5min to play something like a preview.

1 Like

I might be a good idea to clarify what the Resolve OFX plugin can and can’t currently do.

For example, there are no upscaling settings in the plugin ui. So does the plugin only perform x1 enhancement using the models, or can the plugin scale a lower resolution clip to the timeline resolution?

Also, does the plugin have it’s own separate system settings, or does it just take them from the main TVAI program?