Hey, everyone! I’ve been upscaling some videos from 1080p to 4K using Artemis, and I’m really happy with the results. But I recently gave Proteus a try and really loved the results too. The issue with Artemis is that it can look water-colored or unnatural at times. But with Proteus those issues don’t exist. So I’ve decided to begin using Proteus. I just wanted to ask how to go about using it. All the fine-tune settings seem a bit overwhelming. Does anyone have a preset for the best settings that might work for every video (at least 1080p videos)? If not, a general guideline or process to follow when adjusting the settings? Thanks!
1 thing. Use grain.
What I usually do is to first reset the settings to zero, use the auto feature to get a baseline, then adjust both the details and sharpness in 10-15 level increments… as every video has varying levels of detail/quality, there is no real static settings aside from the auto adjustment feature to use for individual videos…
Okay, thanks for your advice!
Why should I use grain?
Realistic + Professional + Hide artifacts
Okay, thanks for your advice!
In the Topaz Video Enhance AI Facebook group, Joel Hruska posted a great how-to for Proteus on Oct. 23rd. If you have Facebook, that is definitely worth a read.
Sorry for replying to you so late; I haven’t been checking this post recently. That post you mentioned sounds very interesting, but I don’t have a Facebook account. Is there a way you could summarize the post to me here? Thanks!
how much typically?
please post the Facebook link in here
Took a while to find the post Richard mentioned. Here’s a copy of the post, as I could not generate a direct link to the great post by Joel Hruska. No intention to plagiarize; just to facilitate the discussion here.
General advice to the group: Use Proteus, but ignore the auto tuner. The auto tuner can be useful, but it does not give the kind of achievable quality boost that Proteus can.
How to Configure Proteus:
When testing Proteus against content, you should use the following settings:
Revert Compression: Between 20 - 75.
Enhance Detail: Test a range between 20 - 75. Above 75 is almost always too much. Below 20, and the improvement is quite small.
Sharpen: I recommend keeping this very low to off.
Reduce Noise: I denoise with other applications and only use this option a little. Recommend settings of 10 - 25, if used. If you perform noise processing elsewhere, you may be able to leave this off.
DeHalo: If you have haloing problems you can turn this up. If you don’t have haloing problems, don’t. Leave it low in any case, it’s damaging to detail. I recommend testing settings of 0 - 25.
Antialias / Deblur: This is one of the most important settings, and how you use it varies by source. You need to test both negative and positive values on your content while holding other settings constant.
For example: Here is a frame comparison between two different Proteus settings. They use the same value for five settings: 60-60-0-0-0-30, and 60-60-0-0-0-(-30).
I’m holding most values at 0 so you can see the impact of the two antialiasing values. This particular frame has a lot of fine detail in it, so it’s a good test for the idea.
So what happens when we vary the first two options? I recommend keeping them similar, though I sometimes let one slide 10-15 points against the other (60 / 75 or 50/65, etc).
This comparison shows 60-60-0-0-0-30 against 45-45-0-0-0-30:
All of the versions of this frame that I’ve showed are promising, but none of them are exactly what we want. One set is too blurry. The other set is a bit too sharp. What do?
Let’s add some secondary settings. Here’s 45-45-20-10-30 compared against 45-45-0-0-0-30.
We get some improvement. It’s not enough. In a few cases, the new settings actually make the problem worse.
This might seem to be a showstopper, but it shouldn’t be. There are two ways to solve this problem:
1). You can spend the time to test Proteus settings carefully, making very fine adjustments. You may or may not get exactly what you are looking for.
2). Start blending in output from other models using an application like DaVinci Resolve. You do not need the paid version to do this.
The comparison below shows 30-30-0-10-5-20 against 60-60-0-0-50 – but the second encode’s output has been blended with the result from a standard ALQ run.
Blending back the data from a different encode makes a difference in the final quality. Here’s a comparison of ALQS blending against Dione:
All of these outputs can also be adjusted depending on your AviSynth encode settings. Sometimes you may want to perform less sharpening or other image processing in AviSynth in order to push more work over to Proteus.
Hopefully these comparisons are helpful as far as output tuning and what the various possibilities can be. Do not be afraid to screw around with Proteus tuning. Try both positive and negative Antialias / Deblur values, and if one doesn’t work, test its opposite (-30 / 30, -50 / 30, etc).
Myth doesn’t work I did it with mutiple levels of grain still gives that oversharp look on certain vids
Hello I tried to PM you but never saw that as a option. This Proteus setting seems to change everytime you move a scene. I am trying to do a Sci-fi movie and the setting change on every scene. I wish the auto would work on every different scene and this does not have that option. Do you think they will ever have that as an option?
I think they fixed this issue in the latest update. You can change it in the settings.
My experience with Topaz VE AI over the past 10 months has taught me that there isn’t a superlative, go-to algorithm. For every video that I intend to process, I usually look at the results under Artemis High, Artemis Medium, Gaia (if its a high quality source) and numerous settings under Proteus. Until I do this and inspect frames at a couple different levels of magnification, I really don’t know what will work best. If you’re getting water-color or plastic textures with Artemis High (the most conservative flavor)), then chances are none of the other Artemis algorithms will work for you.
Proteus is a tricky algorithm. While I can generally tweak Proteus to give a good visual result within the software, I’m sometimes surprised by weird artifacts that are only present when I try to playback on a different device. For example, if I upscale a DVD-quality video to 1080p with Proteus, it may look great within the software. When I try to play back the video on a different device, however, the video has the appearance of being choppy and having too slow of a frame rate. There is also a layering effect most noticeable on skin and hair which looks like overlapping waves, giving the character a blurred or ethereal appearance. Before anyone asks, yes, these are progressive scan videos with appropriate frame rate. This does not always happen with Proteus, but it does often enough that I’m encouraged to use other algorithms when they give acceptable results.
I agree with the user who said that one of the most important Proteus settings is the antialias/deblur. I don’t agree with such a formulaic approach, though. Whether it falls on the left or the right, I generally keep the antialias/deblur setting within 30 points of zero. Revert compression is equally important. You obviously need some revert compression to improve the image. Proteus frequently fails on extracting/enhancing faces that are distant from the camera. For DVD quality sources, this means that Proteus will come up with interesting and sometimes grotesque approximations of faces that makes distant characters look like killer mannequins (the ones that escaped the fire and are steadily approaching with molten faces). Proteus can also have this effect on near facial features, especially the eyes and corners of the mouth. I usually test Proteus through different scenes in the video to make sure this isn’t noticeably happening; if it is, I back the revert compression off.
For a video source which starts off at high quality, say a well-recorded file at 1080p, I’ve had good luck with Artemis High, Gaia and Theia for upscaling to 4K.