Topaz Video AI v3.5.1

Hello everyone,

A new release of Topaz Video AI is now available.

Released September 26th, 2023


Changelog from 3.5.0:

  • Fixes the frame interpolation models repeating frames when changing FPS.
  • Reduces model loading time for Nyx, Proteus, Iris Auto and Relative to Auto.
  • Fixes stabilization not running on lower memory machines.
  • Fixes crashes related to image sequence fps change.
  • Adds warning when rotating after setting crop dimensions.
  • Adds 10-bit for NVENC AV1.
  • Automatic parameter estimation now runs on 8 frames instead of 20.
  • Fixes for Nyx as second enhancement: will now automatically make the first enhancement perform upscaling.
  • Iris-v2 is supported on Intel ARC.

Known Issues:

  • Intel ARC is currently unsupported.
  • For Interlaced and Interlaced-Progressive videos, setting FPS back to Original may cause issues. To get around this, re-import the file. This only happens after manually changing the FPS.
  • Videos with mismatched metadata and streams will display incorrect duration
  • Frame number preview length may shorten on app restart
  • Some vertical videos will be rotated an additional 90 degrees when imported

The known issue contradicts the newly added support for Iris-v2 on Intel ARC. Maybe the known issue should be “Not all models are guaranteed to work on Intel Arc”. Maybe include a list of supported models?


When Proteus will receive some love (new revisions/enhancements) from Topaz again? :sneezing_face: :cry:


ARC is a disaster in Topaz. They couldn’t possibly say Intel ARC is supported yet

When will this get fixed? :thinking:

I was able to set 25fps in older versions, but now’s not possible to interpolated to 25fps anymore :rage:


I reverted back to version 3:3:4 to keep using 50i to 25p.


What does this actually mean? That the auto parameter estimation is worse because it’s getting less input? That it’s better because the processing is more targeted to each frame? That there’s simply a speed increase?

This is incredibly vague and concerning if the results are going to be worse

1 Like

This is only a guess but to me it sounds like the parameter estimation will be the average over a smaller sample of frames so it will be more accurate overall.

Would need Topaz staff to confirm this, though.


what does mat mean? what is “Automatic parameter estimation” ?

1 Like

It’s the automatic values that are set for the ‘sliders’ from frame to frame


Yes! But this needs to be fixed for the future Versions…

1 Like

Not sure what has happened to recent versions, it used to work great and did a fantastic job of converting DV PAL 25fps 16:9 to Deinterlace and Upscale to HD, apart from the 16:9 tags never seem to work nicely when opening in other apps. Now it just seems to get stuck on ‘Downloading Model’?

I have about 50 hours of footage to convert on a feature shot in 2000. I don’t really understand why the template for ‘Deinterlace and Upscale to HD’ defaults to 50fps, if that is the case the template should read ‘Deinterlace, Slow Motion, and Upscale to HD’, and who uses that?

I have to say I’m mega impressed with the deinterlace and upscaling output, when it works.

1 Like

Is this just the Beta unchanged release version?

1 Like

Reverted back to v3.1.0 to test and apart from being completely out of whack on export size, ‘Downloads Model’ fine. Is the ‘Downloads Model’ not playing nicely with Apple’s Private Relay option in newer versions?

On the whacked out export size, should not DV PAL 16:9 to HD not be a straight upscale to HD with is the same aspect ratio of 16:9? See photo uploaded on what seems to be happening with v3.1.0 and any other version I can get to work.

not to forget:

1 Like

it means and is all of the automatic settings were not so automatic it was on 20 frames per second than did the automatic stuff averaging… this number was a hidden feature once emailed by support…
now if you have 25 frames in a second a shit lot ca happen creating artifacts
now when we do it by 8 frames per second the automatic will be more precise hence this is a great enhancement

The results I now get with my preset look awful. If I define a specific model for the second enhancement I do want it to be the second enhancement and NOT the first.

This is a nonsensical change.

60fps videos generated with apollo v8 are being generated corrupted, on 10seg of apollo video.

original 30fps video and apollo v8 60fps zip (14.4 MB)

1 Like

and in English?

This change does not force Nyx as the first model to run in a Second Enhancement mode. Nyx currently only runs in 1x scaling mode, so in this case the first model (Iris for example) will run at either 1x, 2x, or 4x scaling mode while Nyx will just denoise without additional scaling.