Tip: Three step tuning method, to recover bad quality videos

How do you upscale using NNEDI3? with what tool?

The files I have sent are progressive, not interlaced. So for the time being I am focusing on these old AVI files. But thank you for the comments. Maybe soon I shall also ask about deinterlacing, since I have some “i” footage, which by the way is wrongly recognized as progressive…

Selur’s Hybrid.
PS What do you think about my result? Maybe , after cleaning footage, I could use Proteus with some tweaks.

I like it, that is why I asked. if you could post your Hybrid settings for your upscale (NNEDI3, etc.) & cleaning I would love to give it a shot on my stuff.

In general there is a mixture of Dehalo Filters and Denoise ones. I used to try FFT3D one, KLN, QTGMC. It is really a problem to find out really good solution. I do hope it will help you somehow. In the case I just use NNEDI3 default setting.

Get rid of compression - tip

“Fix Compression” is well known, it resolves compression artifacts, which are nothing other than break down pixel clusters. However, this leads to more bluring and too high values make the image look unatural, which is the main problem. But it’s good and usefull, well (low) dosed!

Concrete: If you use “Iris” model, revert compression values above 5 are usually too high. But you can use “Dehalo” additionally to smooth out clusters. Unfortunately “Dehalo” also leads to undesirable effects, if value is to high.

But there is a third method most don’t use it in this way, but helps a lot: Additional to “Fix Compression” and “Dehalo” you can use negative Values of “Anti-Alias/Deblur” yes most people use positive but if you have compression this could be wrong, then use negative value.

Example of a combination:
Fix Compression: 3
Dehalo: 1
Anti-Alias/Deblur: -3

I use Dehalo filters, YAHR and FineDehalo (try different settings depending on the footage) and absolutely different Denoise filters with different sucess or unsuccess,…now the question is how to upscale to get much detailed footage…

A distinction must be made between physical or visually perceived details. Even it sounds strange, Image enhancement is more or less accompanied by loss of information, especially when you denoise, but it brings improvement for the viewer in other way, because what is there becomes (more) visible.

Fix Compression, Dehalo and Anti-Alias/Deblur do simmilar at different ways. All of them changing the differences in brightness between neighboring pixels. Result ist either more depth is created, or the opposite. Optimal values ​​are very individual depending on the source.

I always look for dark areas in the image and play with the three parameters until large collections of pixels of the same color dissolve by becoming more graduated. But when a brightness difference between pixels is very small, is not good either, then you have contrast losses.

One of the most important things is denoising, you’re talking about. Whatever is a compromise between loosing and keeping Image Information. Unwanted noise should disapear but the software don’t know what should go away and what not, you have no way of marking areas in the image to denoise individually. I hope Topaz Labs puts more focus on denoise. NYX is a good approach but that doesn’t really convince me yet, I think there’s still a lot of potential.

1 Like

Where is the IRIS model, I only have Proteus, Artemis, Gaia and Theia

There’s an Nyx 3 in alpha testing right now.
I agree that denoising is one of the key enhancements that needs to be done well… but, Proteus 4 can do some amazing things with just ‘Fix Compression’ (it denoises), ‘Recover Details’ and ‘Sharpen’.

I have also noticed this lately that Prob4 Fix Compression also does a pretty good job denoising with not much detail loss compared to denoise only.
so I play around with “Fix Compression” and “Improve Details” to find a sweet spot.
as more “Improve Details” value you use, the sharper the footage would be. if you use anything above 30-40, I don’t think you need to apply Sharpening at all, you can skip it, leave it at 0.

do you have old version? Iris model V1 was launched Version 3.3 then later Iris V2 was released. Iris v1: Face Enhancement for Low-Quality Videos (June 2023)

“Fix Compression” and “Improve Details” dos not really denoise, but it seems like denoise because noise blends in with the rest. That could be desirable, but the problem is “Fix Compression” is a form of bluring, high values leads to unnatural look. “Improve Details” reinforces structures, high values leads to artefacts. You can see this often when noise and ears of a person look disfigured.

When I started with Topaz, I used often high values ​​of the two functions, up to 25. Today I use “Recover Details” values ​​below 5 and “Improve Details” below 15, when using “Iris” model even often only 8-12 of “Improve Details”. (relative to auto)

I just found out that there are other ways to get an image sharp and create deepness, good contrast and shiny look. As I described in my prevous post this three functions are important doing that, investing a lot of time to find optimal values ​​and doing good denoise. Then I do a second pass as finisher, using “Artemis HQ” or in some case the harder one “Artemis Strong Halo” or “Theia Fidelity”


if anything, i noticed that reduce noise will also fix mild compression, and will do so with less blurriness than using fix compression.

as for giving the footage a more 3D look using artemis strong halo, i found its the same thing as relighting a scene, and imo its better and more efficient to do so in davinci, using qualifier on the luminance highlight and up the gamma or gain a bit. this also doesnt produce unwanted artifacts or unwanted smoothening of the skin, and is consistent over the whole scene. u have full control over how strong the effect too.

I love “Strong Halo”, ok it’s difficult and not for every video a good choice. It’s stronger than all other models “burn in” effect, like when you use furniture polish.

ya, like need to get lucky and pray “strong halo” wont produce artifacts, but if doing it in davinci it will work 100% every video.

you can “prepare” it for “Strong Hallo” at first (main) pass, by taking more care as usual getting no artefacts.

Then before using “Strong Halo” as finisher, you can use Proteus in manual mode (NOT relative to auto): Set denoise to 3, and do nothing else. Then click “add Second Enhancement” and choose “Artemis Strong Halo”

The very slightly denoise before Artemis leads to a more shiny result :slightly_smiling_face:

@Mayday Yes I have version V3.1.11. How can I do without the Iris model?

i upgraded to V.4.0.3. Comparing my results of just using proteus Auto and nothing more, i dont get same results. Do you know the parameters of Proteus to get same results as in V3.1 ?

Which Proteus Pro3 or Pro4?