Question about Nyx

I have some high resolution footage that was shot under poor lighting conditions that only needed noise removal. But when running experiments with Nyx 1, it seemed to remove quite a bit of detail in shadows. Has anyone else observed this and if so, have you found a satisfying workflow to avoid the problem?

I love Topaz Video AI. The models do magic. But I haven’t found the “magic model” yet for many of my situations…

  • Proteus is razor sharp and usually good at noise removal and bringing up detail, but often does terrible things with faces and sometimes creates weird texture effects on skin when it doesn’t properly remove noise.
  • Iris is kind to faces and good at noise removal, but tends to drop detail and greatly soften imagery
  • Nyx sounded promising, but again the aforementioned problem.

Don’t take any of this as complaint. I know the models are a work in progress and the software is still light years ahead of other options.

2 Likes

Try Dione TV and Dione DV. they clean up pretty well, but far less aggressive then the others.
To use Dione AI on progressive videos read here

I haven’t found a way to denoise without losing some detail yet. And that might just be the nature of denoising. Nyx 2 was an exciting idea, but currently over-sharpens.
I have not tried all the ways of denoising offered in Hybrid, just EZDenoise. I can get near similar results with Nyx 1 manual, just denoise set from around 15 to 30. The rest zeros.
Over the last few days I’ve been trying out Nyx 1 very thoroughly. It kind of comes down to the old finding of: To make an exceptional enhanced video, you would need to make settings tuned to every scene.

People keep saying great things about Neat Video, but it looks like it’s got no command line interface option, so that’s out for me. I’m not going to manually create profiles for each episode in a TV series when usually what works good on one episode, works good enough on all.

I used Neat as my go-to for noise removal before I discovered Topaz. Neat is far better than Adobe’s noise removal tools in Premiere and After Effects and works great for very light noise that’s somewhat consistent in a clip, but it can’t handle heavy noise or noise that varies a lot in a clip. That’s where Topaz truly shines in comparison, despite its other shortcomings.

2 Likes

That’s comforting to hear… I suppose it doesn’t help with your problem. Maybe it’s a colorspace conversion thing?

That’s very helpful. I also use Hybrid to deinterlace videos and thus viewed Dione as irrelevant.

Speaking as a graphic designer/video editor that doesn’t write code, I really hope Topaz will integrate your suggestion about creating a setting. Editing scripts isn’t in my comfort zone.

I used neatvideo a lot before veai came out, and I still use it to refine my videos. as you say the noise must be constant for neat video. I have scenes in my videos where I cannot remove certain artifacts without losing too much detail whereas with qtgmc + artemis hight/medium, the artifacts disappear while leaving more detail than with neat video. qtgmc (ezkeepgrain: 0.5) + artemis hight (recovery original detail: 100) + iris medium probably gives me the best results so far. we could even do without artemis if we define “final temporal smoothing: 1, 2 or 3” but in my case I always leave it at 0, because on the movements it loses too much detail, and it seems unreal once the upscaled and improved video

When you run the Artemis High as pass 1, you keep it at the same resolution or you upscale with Artemis and then just do an Iris v2 cleanup?
which one you do the Upscale with? I want to try this.

Is Neat video a stand alone tool/program, or it is only available as a plugin to an existing commercial tool?

I use Artemis Hight at 100% (720x576 or 768x576 square pixel) or Artemis Hight + another model which removes a maximum of artifacts because then I use Iris to upscale to HD (the secret is to be able to have a clean video before upscaling, an SD video without artifacts and with maximum detail, which allows you to use iris by being able to increase details, dehalo, sharpness without increasing artifacts because of upscaling).

neat video is a powerful plugin, but it may require more resources than vai. I have almost never used it alone with Vegas, but always with dozens of filters. and when i preview or produce with vegas (hd) i am at 3 or 4 fps (gtx 1660, ryzen 7 2700), i think with neat video alone i can be at around 15 fps) but neat video is a plugin available for these software:

on the other hand neat video is quite hard to configure well when you don’t know, and everything is based on a noise profile that you define by making selections on parts of an image (light tones to dark tones) so it’s is a little tedious, a profile which is displayed in percentage, and the higher the definition of the video, the simpler the creation of the profile, with an SD video, it is quite complicated and it is better to define several profiles depending on the scenes , I rarely reached 100% on a profile with an SD video, I achieved it on part of the video with a clear sky between blue and orange (for the selection of the frame to apply in the image to create a profile, it must be green and it takes up a good part of the image (here a basic profile at 61%):

look here on this scene, the frame does not pass over a totally uniform area, so not great for properly analyzing the noise so it warns that it is not uniform in the luminance:

here, the same scene a few images later, the frame passes, it shows me a good profile (66%) it’s quite good and a little better than the first scene with its 61% because there are quite a few light and dark areas. there are a few points in the red green blue areas that we can complete (the more points we fill in manually afterwards, the better the percentage of the profile):

here, the same scene as before, with the same selection but tweaking the missing points, (it goes up to 75% because I filled a few squares, we fill them by selecting small areas in the image (smaller than for make a basic profile) on the right, circled in red, this is the quality of the profile, circled in green these are the points that I cannot complete because it goes quite black in the RGB range, and not white enough either . the double arrow in black is the color intensity curve (from dark to light tone):

now here’s why it’s more precise when the definition of the video increases (the frame goes into a slightly larger definition and given that the noise is therefore a little finer, the profile is basically of better quality):

on this scene for example, even if HD the frame is in yellow, therefore too small for analysis, which gives a bad profile (46%) and which will lead to too much loss of detail:

That’s basically the creation of a profile with neat video, then there is the whole part of the settings on temporal and spatial denoising with various options, and quite a few options, which makes neat video still powerful , and when we have a profile set at 75%, (hd) in the case of this scene here, here is what it gives (mix with the original at 75%):


(with a profile set to 100% there will be less loss of detail) so here is this profile that I made is only at 75% but it gives an overview of the power of neatvideo, it denoises very well but it does not remove the large artifacts that can be seen on the edges of the boat’s windows. or to remove them, you have to push the sliders which leads to losing more detail

2 Likes

wow, thank you for the detailed explanation.

No problem. :slight_smile: this is only part of the neat video settings. but I think that everything is better explained on their site anyway. The advantage of this plugin is that you can save different profiles and settings. So easy to change and find your way around when you want to change settings on different scenes

Another challenge with Neat sometimes is getting an optimal noise profile in clips where there are very few (or very small) “flat” areas without detail. Again, TVAI shines in this regard. When I first started using TVAI and was having issues with Proteus weirdness, I tried using Neat to clean videos first before upscaling and sharpening with TVAI. I still had better results with TVAI.

Like your example, I find Artemis or Iris are the best models for achieving my result, albeit with some compromise.

1 Like

I agree :wink: neat video is good for final process for me.

so the question is, if I own TVAI, would there be any real added value getting neat video or most of the things are going to be covered or in many cases better in TVAI?

I think it depends on the footage you’re usually working with. If most of your footage is HD and UHD and all you need often is some noise removal, Neat is probably a worthwhile investment. If your footage is often poor condition, low res or has compression artifacts and complex noise problems, I wouldn’t bother with Neat.

Personally, I haven’t used Neat much since I discovered TVAI.

1 Like

I’ll ask you this: when you have a noisy video, what tool you head off 1st to treat (or attempt treating) it now days?
TVAI or Neat?

TVAI. If the noise is light and consistent, I may try an experiment with Neat. But TVAI is where I land 99% of the time.

1 Like

and even if neat video uses the gpu and the cpu at the same time, vai is faster in my opinion. but indeed neat video will always lose detail, especially if the noise is not constant and especially since it can improve sharpness, and smooth edges, remove some slight artifacts, but not as efficient as vai at this level because neat has no dehalo or compression, it’s not even based on AI :wink:

Exactly. I only use Neat now for very pristine video where all I need is some basic, light noise removal and the risk of TVAI getting too creative exceeds expected benefits. That’s not often these days.

Whenever I do use Neat, I usually end up using an unsharp mask effect to compensate for softening.

By the way

  1. what is your Iris settings you are using for the upscale after you use Artemis High Quality?
  2. you using Iris v1 or v2?