Topaz Video v1.1.0 - DVD upscale HD settings recommendations?

Just started to use Topaz Video v1.0.4 and upgraded to v1.1.0 for upscaling my DVD collection to HD.

Been going back and forth for the optimal settings and consulting ChatGPT as well and ended up with following settings for my kungfu action DVD collection. My results are satisfying but i wonder what other people use.

I tried Dione TV, although this gave good results but the face skin textures in dark scenes is bad.

So i ended up using Proteus model with more control and the settings:

  • HD 1080, Interlaced
  • Recover detail: 15
  • Reduce noise: 24
  • Sharpen: 4
  • Dehalo: 10
  • Anti-alias / Deblur: 12
  • All other settings off.

Export as Prores 422LT without audio.

Afterwards i use handbrake, h265 10bit, slow, RF22, main10, encoder tune Auto with flags

strong-intra-smoothing=0:rect=0:aq-mode=1:rd=4:psy-rd=0.75:psy-rdoq=4.0:rdoq-level=1:rskip=2

to encode it to mkv and use MkvToolnix to combine it all with audio/subtitles/chapters.

Some Blu rays (old 80/90s movies) i tried had a lot of filmgrain, which resulted in big files after handbrake. I tried to remove the filmgrain with Topaz Video, but that would take an insane time of 11 hours on my RTX 4070 so i have left that as it was and accepted the large mkv file. I think without the filmgrain, the handbrake file would be compressed more efficiently.

Like to hear what other people are using as their settings for DVD.

I think Proteus is the best so far.

  • Fix compression: 20-50, this is the denoiser.
  • Recover detail: 12 max
  • Reduce noise: 0, I only remember this ruining the picture, not denoising.
  • Sharpen: 0-18 max
  • Dehalo: 0, this blurs the whole picture.
  • Anti-alias / Deblur: 0-24
  • All other settings off.

Sharpen and Anti-alias / Deblur tend to do the same thing, If Sharpen overdoes it, then use Anti-alias / Deblur.

For denoising Blu-rays, use Nyx 3 and set everything to 0 except the first option. Turn it up until the noise is less or mostly gone. You can also use Nyx Fast. It gives similar results but at different slider values. The trick is to find the scene with the worst/most noise and maybe not remove all of it, but reduce it to a level that’s still not distracting. You can easily go too far an it starts to wad all the noise into clumps. That’s very distracting.

CRF 22 is probably okay for most videos, but I have found that 18 keeps things cleaner with not too much more file size.

Thank you for your advice. Much appreciated. I will try that out.

I wasn’t aware of other models in my interface, but added Nyx now. Where is Nyx 3? I only see Nyx, Nyx Fast and Nyx XL that can be added.

I have the best results for older DVD sources with using Iris LQ V1 and custom settings (that vary a bit on the source).

Iris does a really thorough clean-up of typical issues of older sources, especially (chroma-) noise, also partly Color-banding and moirés, also those often really heavy Halo’ing found in those videos - and can give quite a bit extra details/resolution. In contrary to the newer versions LQ V1 only quite seldom does strange things (albeit the occasional „monster face“ in the background might appear).

I found Proteus is the best for me with the following:

  • Fix compression: 30/40

  • Improve detail: 30/40

  • Reduce noise: 0

  • Sharpen: 0

  • Dehalo: 0

  • Anti-alias / Deblur: 0

  • Recover Detail: 20

  • All other settings off.

30/40 = if you set 30 at Fix compression, set also Improve detail at 30. same goes with 40.
I have noticed they go well together.

What I have noticed that Fix compression , although removes noise very nicely, it can remove details (not as badly as Reduce Noise), so the Improve detail, restores those lost details and also it does sharpening/deblur at the same time. so you don’t need to use sharpening and/or deblur at all when combining them both (Fix compression/Improve detail).

2 Likes

Hey! I’ve just gotten back into working on my old SD videos! :slight_smile: I tried your settings, so I’m testing them on a deinterlaced video with Muksun in VirtualDub, because I still find it to be the best for now. I tried your settings with Proteus (dynamic, not manual). The video is 700x564 (I cropped it to remove the black borders), and I accidentally tried x1 (700x564), it’s really not bad! On small faces filmed from quite far away, it doesn’t distort them, and there are no artifacts; the details are well preserved! I then tried x2 (1400x1128), and it looks pretty good too. I’ll try to fine-tune it :wink:

1 Like

Nyx is Nyx 3 by default. You can enable older version of all models in preferences. You’ll need to do that if you want to try the suggested Iris model.

I am also going to give these a test. I should have done them before, but I might not have. There’s only a large amount of possible setting combinations. :stuck_out_tongue:

1 Like

the settings that I posted are Manual, not sure how those would react when used as Dynamic.
All my sources are 720x540 , x2 upscaling, i.e 1440x1080.

Once you fine tune it, please post your settings..

There are two worlds; the old good one Proteus OR Starlight Mini & Sharp. So Proteus is solid but the image improvements are not spectacular. The new stable diffusion based ones are high reconstruction models, could be promematic based on content but when you have license for local use (and the gpu hardware), you must try it.

Workflow to try

  1. Bring your source to 540p vertical lines using simple Lanczos, NTSC 480 → 540 for PAL 576 → 540. You can use hybrid, Virtualdub or something else. Here you can also crop a little bit when needed.

  2. Model Starlight Mini 2x Upscale 540 to 1080

  3. Reimport result into Topaz, choose model “Starlight Sharp” and do 1x no upscale. When you do 1080, Starlight Sharp does not use built in NYX-XL and the result is much more natural. Both SLS and SLM does some denoise, by how the model works.

1 Like

Though I’m sure the results are more substantial, I cannot recommend them because even with a 5090 it takes days for a single video. Also, did they ever fix the expression change in those models?

My initial test of those settings is positive. I must not have ever run Improve detail without Sharpen. Anyway, I’m trying some more challenging DVDs with that.

1 Like

Well, you got the baseline settings/Idea, now all you need to do is tweak a bit those two values to hit the golden spot, but try sticking with those two values only leaving the rest alone (at 0), Apart from Recover Detail that i like keeping at 20, it helps maintaining the authenticity of the Original picture.

NOTE: the higher the Improve details, the more sharpening it applies (but less distortion than regular sharpening filters), bear that in mind.

I have noticed those two work pretty good together, but only in Proteus. on other models like Iris, they don’t yield such impressive results.

Manual mode is what I use, not Dynamic, as Dynamic will add other filters in the background.

You can try another trick.
bump the Recover Detail to like 40-70 (play with the numbers, but the idea is High value), and then apply Fix compression to like 40 or 50, that will keep a lot of original details, but less noisy. I don’t remember if I used Improve details on that trick. If you do add Improve details, it might emphasize the original flaws of your source, as you on High recover detail settings, so try with low numbers, if you decide to use the improve detail as well.

That will really keep the authenticity of the video, the vintage look as the source, etc., just double in size

I figure Recover Detail will more likely preserve what I’m trying to have removed or fixed, so I keep it low or off.

Improve detail seems to have worked pretty well on my more extreme test. I think If I can pare it with a good dering filter, I might have what I’ve been looking for. The results won’t be as good as a quality Blu-ray source, but they will be easily better than the DVD.

1 Like

With those settings I posted, I noticed those work less good at x1, they are more intendent for x2.
I did not try other upscaling size other than x1 and x2.

You are correct in general, but I have noticed that Recover Detail, let’s say at 20 or 30 max, helps restore the original details that were lost due to Fix compression.
Recover Detail is less good in case of noticeable noise / scratches at the source (it might leave subtle noise, but are perfectly fine if the video has mild-moderate noise or is just at lower quality or with rough edges.

I ran a quick run on my test video with
Fix compression: 40
Improve detail: 40
Recover Detail: 30 - 40
And I notices more details were preserved as opposed running the same with Recover Detail at “0” and I hardly spotted any noise left overs (when pausing the frame).
I use Video compare to spot quality differences, as it is some time really hard to spot differences when running a video one right after the other.

“Recover (Orig) Details” default value Topaz set is 20, does something like a “glossy brochure”, it brings more shine but tends to waxy by smears “pixel blocks”, you loose some finest details. However, it cannot be said in general terms whether this is good or bad; it depends on footage you have. (and “fix compression” I try to avoid because it’s invasive)

1 Like

Ah but “fix compression” is the enhancer within Proteus 4. Of course, don’t use it if it’s doing undesirable things to your videos. From all the DVDs I have processed, I usually like what fix compression does to them. So far, there are only two TV shows I have that I could not find settings that looked good with Proteus. They are older with heavy grain. In the darker scenes I could not get it to produce an image that did not have distracting noise.

I simply noticed that the built-in models a they was trained, so model built in “fixed compression” without user parameters, are the least invasive. As soon as you set one individual parameter, this changes, even if you only set the value to 1 big things happens. So either use it out of the box, i.e., in Auto Mode, or if you use it, then high values to achieve a desired effect, but it’s bad if, for example, you set the value to only 5, then a pixel shift occurs, but the benefit of the function is hardly noticeable.

With Proteus, it’s not so pronounced, but with Iris LQ, parameters has an extreme distorting effect on the image when you adjust user settings. With Iris LQ, I would only use “Deblur,” which is the least invasive and very useful because model is so soft; even the sharpness control in Iris LQ has a dramatic effect. Interestingly, “Iris MQ” is quite different from LQ, similar to Proteus, and can also be used with high values to “de-noise.”

We’ve noticed that the Fix Compression feature performs exceptionally well in Proteus.

In Iris, however, it falls short. The reason Proteus’s Fix Compression stands out is that it acts as an outstanding denoiser with virtually no image smudging (like a layer of vasaline) or artifacting—unlike most other denoisers in Topaz. This unique behavior is exactly why we prefer it so strongly in Proteus.

In Iris, the feature simply doesn’t behave the same way, so it’s not advisable to use it there expecting the same superior denoising results.

1 Like