I share my image enhancement experience with Davinci Resolve, Hybrid, Vai and Vegas pro

hello all, I have the solution for “themis deblur” with the checkerboard which works well;) and I made a clear advance in my quality settings that in my opinion no one has to test. I wanted to try the deinterlacing again with “davinci resolve”, because despite QTGMC being superior, it is still the second best, the thing is that it deinterlaces very well despite having some jagged edges but barely visible, most some time it smoothes the edges very well but on the other hand it does not remove noise like QTGMC so what I tried was denoising with “davinci resolve” on the same timeline and the same deinterlaced video, I applied it slightly like this: temporal: best quality, large amplitude: 1 frame: 10%: movement 50%, statial: Improved (On the picture it is on “better quality” but I discovered after that we can put on “improved” for better final results) : small radius, luma and chroma: 1%, it removes the biggest visible noise without removing details even in movements. (view the photo below, difference not very visible because a slight noise reduction is enough, but that’s enough) Then I export in prores avi uncompressed (sd pal) 50 fps.

(we do not clearly see the real quality of the photos posted due to the compression that the site does to upload the images, I think.)

Once finished, I import this video into “hybrid”, I apply the “MLDEGRAIN” filter, two dehalo filters, (“dehalo alpha” and “blind dehalo”, I have my settings but it all depends on the source of the video, like for the noise) , I activate “QTGMC” with “EZKEEPGRAIN”: 3.00. The “line” filter “NNEDI3AA”, and the filter “FILMGRAIN”: 0.02, then I export all that in “prores 4444 XQ”.

The third pass is to use VAI, from there you will see that each model denoises much better and leaves more detail than before. I didn’t use proteus before because I couldn’t adjust anything clean, now I can much more easily, it’s a pleasure! In fact the video output from “hybrid” is so clean (even if there is noise) that there is almost no detail difference between “artemis hight” and “artemis medium”, so “artemis medium” is superior because it deblur better than hight and makes the edges a bit sharper. I tried “artemis low” too, we lose a little more detail (I think invisible when reading) but we still blur the image a little better. Then for “proteus”, it’s now that I understand the potential of “proteus”, we can recover details without increasing the artefacts, so the image can seem more realistic, while playing with dehalo, deblurring, denoising etc, we rarely loses detail even pushing the sliders. I export all that in “prores XQ 4444” of course.

Then I import into vegas pro to finish tweaking my settings, in order: a “gaussian blur” filter, “neatvideo”, a “magic sharpen” filter, a “detail recovery” filter, a “dehalo” filter (dv fixer), a “color adjustments” filter, and finally a “grain” filter.

So I would basically say that thanks to “davinci Resolve” and its deinterlacing with its noise reduction, we can do more miracles.
Of course, the settings will depend on the source videos, but in my case, they are dv pal videos digitized in firewire dating from 1996 to 2002, I think a lot of users are having fun with the quality of their videos right now of family/memories etc. I’m very picky about quality, it’s been 7 - 8 years since I’ve done a single video editing but only tests on a reference video of 1h15. And for once I think topaz have made a good software to manage this kind of video, I’m still very satisfied despite the slowness of v3 (in my case), that’s why I tested everything these settings mentioned before with 2.6.4 which is fast (0.16 spf)

I tried “themis deblur”, which is not very good with my current settings but the solution I found, which works well in my case is that once the video is imported into vegas pro, the first filter I matters it’s a “gaussian blur” usually it is settled to 0.0002 and 0.0005 (just to smooth the edges) but in the case of a video with a visible checkerboard due to “themis deblur” setting it between 0.0009 and 0.0011 in SD and 0.0006 and 0.0009 in HD or higher removes this checkerboard, with the second “neatvideo” filter. after adjusting this one, then you have to play with the sharpness filters, bbc “magik sharp” etc to regain sharpness, from there, more checkerboards visible thanks to “neatvideo”. (vegas pro filters in order: view the photo below) (in the preview screen, the left part is the video output with topaz, the right part is the video modified with the vegas filters)

So maybe some people have tested Resolve’s “neural engine” deinterlacing and didn’t like it, in my case it handles deinterlacing well, the same for noise reduction, I’ve read on some forums say that it can be as powerful as neatvideo, it’s probably true under a certain noise threshold, but beyond a certain threshold, to know neatvideo by heart, I think this one is better. But neatvideo is much slower and the fact of not removing all the noise from the video but the “biggest” with the noise reduction of Resolve, makes it really practical compared to neat video and surely more powerful, the worst is that I found very good settings on a whim, I only did a test with resolve with the values mentioned at the beginning, but it is possible to test other values with Resolve, and even with Hybrid, that’s 2 softwares to adjust before switching to VAI so a lot of tests still … But what I find remarkable is the final quality and above all, much less blurry details in the movements so much less loss of detail. (view the photo below).
And then with the artemis models, I stay in sd (720 x 576) because I have more details, and it’s just in vegas pro that I put the project in HD, but for Proteus, I produce with size x 200 so 1440 x 1152. (view the photo below, we can clearly see all the water flowing from the fountain whereas before with the noise reduction settings of “hybrid”, coupled with “artemis medium”, it suppressed a lot of water)

Thanks for reading, and really if you have time to try this method, do it! even if you are processing progressive videos, just try to denoise a bit with Resolve before switching to VAI or Hybrid + VAI which can be even better. :wink:
If you have questions… (sorry if it’s my french is badly translated)

ps: the quotation marks are there for the important terms and especially to translate well into French


Thank you for posting this, very helpful!

1 Like

Thanks 4 posting this. Unfortunately i can’t test Resolve deinterlacing because it does not detect my Rtx3070. :thinking:

1 Like

It’s weird, I won’t be able to help you because I don’t usually use Resolve, have you respoiled in the settings? Can be an option with GPU acceleration?

1 Like

Thank you! For my SD videos, I have better results with the denoising options with 1 image for the temporal, the wide amplitude, Luma: 8 movement: 40 and average radius for the space with Luma: 8 :slight_smile: It makes me lose much less detail, the image seems less “plastic” it’s like the source video with more details and less noise with the advantage of producing good results with a VAI model.

I did tried several things even installing the latest studio drivers and no luck. I’m using the demo version so can be because of that. I will try again tomorrow.
Thanks :+1:

ahh it’s possible, on the other hand I find that it’s a hassle to manage the deinterlacing of resolve, you must first set the project on “process deinterlacing” and activate “neural engine” then import a video, after “attribute settings” and switch to “top frame” (in my case)