Topaz Video AI v4.1.0

Hello everyone,

A new release of Topaz Video AI is now available.

Released January 16th, 2024

Changelog from 4.0.9

Model Updates

  • Proteus v4
    • Better handles noise and compression artifacts.
    • Works on interlaced and interlaced progressive videos as well.
    • Improves faces better than Proteus v3.
    • Reduces ghosting of the previous frame at scene changes.
    • Better results on Auto.
    • More temporal consistency.
    • Faster than Proteus v3 in most cases.
  • Iris LQ v2
    • Improvements in detail, especially, for non-face inputs.
    • Improves temporal consistency (reduces 'marching ant' artifacts).
    • Reduces ghosting of the previous frame at scene changes (same as Proteus v4).
    • Reduces color distortion.
    • Adds TensorRT models for faster inference on certain NVIDIA machines.
  • Nyx Fast v1
    • Improvements to processing speed -- twice as fast as Nyx v2.
    • Reduced oversharpening and texture artifacts compared to Nyx v2.
    • Allows tweaking Sharpen slider value (Nyx Fast only).

    Full changelog

  • AVX2 requirement removed and replaced with a warning about possible errors.
  • In a future update, auto-parameter estimation will be moved to the GPU to handle cases where Video AI is running on an incompatible CPU.


  • Rebuilt EXR + DPX support in-app
    • EXR and DPX inputs will now be correctly downsampled to 16-bit TIFF for processing, and exported at 32-bit linear color.
    • This function requires a large amount of available disk space on the drive selected for exports.
    • Support for ACES and other color formats is still underway, and our team is actively seeking different use cases for EXR and DPX to further improve compatibility.
  • Added focus fix feature to enhancement.
  • image

    • Improve missed focus in videos with the new "focus fix" mode for the Enhancement filter.
    • This feature changes the way models affect input videos by downscaling inputs before upscaling.
    • Two different modes can be used to control the strength of this effect:
      • Standard mode first downscales input media to 50% and then upscales to 2x.
      • Strong mode first downscales input media to 25% and then upscales to 4x.

  • Added Full Disk Access prompt to resolve disk access errors on macOS.
  • Improved some UI elements in filters panel.
  • Fixed incorrect frame interpolation related bugs when applying presets.
  • Fixed bug where SAR information was not retained across inputs.
  • Fixed bug where switching between Relative to Auto and Auto modes would not reset sliders.
  • Fixed bug where Nyx is resetting "Recover Original Detail" to 0.
  • Fixed window display related errors.
  • Fixed fullscreen UI issues.
  • Fixed bug where the app would hold a filesystem lock on input videos after app close.
  • Fixed bug where audio codec getting reset during trim change.
  • Fixed bug with metadata loss on MKV exports.
  • Fixed bug where Export Error Resolver tries to resolve preview related errors.

Known Issues

  • Batch processing selection can overwrite FPS output.
  • EXR/DPX Exports may lose input frame number.
  • Inconsistent “Preview X frames” enable status.
  • Rotation w/ previews will intermittently be over-rotated.
  • Looping previews has a stutter.
  • Live preview can have frame de-sync.

Looks really good, a very rich update. Thanks


Nice meaty, juicy update! Will Iris v2 produce better face results than proteus v4?


How does Proteus V4 comapred to Iris medium quality? What does proteus v4 do better and vice versa? Need direct comparison to determine best use cases for both


This is exactly what I’m wondering. I will test myself, but I would like to have some info so I know what to expect.

Nyx Fast model download seems bugged. If you go to the download model section and you try to download it, it will error out.

Portrait orientation video appears rotated and stretched in the application preview window though saving the processed video shows the correct orientation.

Really hoping for a solution on this. Hundreds of videos I’m processing and selecting the framerate manually is a huge pain.


Thank you for AVX2 requirement removing !


If you set the input to hd and change it to 4k, the output will not change to 4k


The newest version 4.10 will not allow image sequence loading of JPG files, but works fine for PNG.
The Browser lists jpg as a valid image input file.


Did old presets get removed? After updating they are no longer there for me. I can still see the files if I browse to them but it seems the format was changed and there is no way to convert them to the new format?

1 Like

@Moebius @josephgomes619
Proteus does a better job on other elements in the footage that are not close up faces then Iris and better job on faces that are not the focus of the video then other AIs (Iris might do a better job in some cases).
Iris remains the main AI for face recovery if that is the major of the video that needs fixing / upscaling.
Proteus 4 seems to be multi purpose enhancer that tries to retain details as much as possible and do much better face job then other AIs including face (except Iris which is more face oriented) and also on any other scenes that are not a face.
So if you have a video which is not a face close up as a major scene , proteus would probably be your AI. as it would perform much better on other stuff then Iris.


This is quite annoying. Noticed same last night.

no it is not. it is shown the input size where you made the mark!

No, it’s not input. The output hasn’t changed

ah, I see. never mind. :smile:

Very nice update, I can safely say I’ve seen notable improvements using Proteus in terms of processing speed at 2X (from a 1080p source) (benchmark results concur as well).

When will the sensible Auto Bandwidth be restored? I don’t know whether it was using AI to approximate or was just a default compromise, but it offered the best compression with near magical quality. Usually within 10-20% both up or usually down, capacity wise.

As it stands, Medium offers such massive bandwidth compression, that I cannot believe it offers the same level of quality and High is a redundancy nightmare bloating the sizes to over twice as much for no discernable difference.

Artemis High Quality is the best AI model for both FHD and UHD Blu-Rays, both to 4K and 8K and I recommend it to anyone whom wants this program to enrich their movies and series collection like I have.

Until AUTO is returned, I see no reason to renew my license ever again, even if Artemis somehow gets a new update after 2 years of stagnation.
As far as I know, this is the last patch containing the Auto Bandwidth function and the quickest: Topaz Video AI v3.2.2. Any version beyond it adds at least a full hour to the wait times for no reason whatsoever. Tested on both an RTX 3080 Ti and an RTX 4090.

Thank you for making this program. It was perfect even after you added +1 hour to wait times, because you soon also added audio conversion to AC3, but removing that perfectly balanced Auto Bandwidth function completely ruined this software for me past 3.2.2.


It’s as I said before, what a magnificent turn-around from the GR 4.0!

I’m thrilled to see the trajectory this prog is going!


Why is the UI getting laggier and laggier the more previews have been created?

Especially when working with 4K files, I notice that many UI interactions like selection playback location, going to next frame, starting a new preview, …, only start after some lag (even wait cursor might show up). The lag time seems to be proportional with the number of generated previews. It get’s pretty obvious after maybe 10 previews. After clearing up the previews, the lag is gone until more previews are generated again.

System: AMD 7950X, 4090, 32GB RAM, Win10