Hi there. I have done many video enhacements now and here is my experience especially in terms of improving bad video material. Often when you use "Proteus and manual values and then set upscale factor, you get artefacts, like a “grid” over some parts of the video. To Avoid this here is my recommendation to get high quality results by doing a three step tuning:
Good denoise is usually the most important basis. Start with “Proteus” and set it to “Manual” then looking for a place in the video that is sharp and rich in contrast, also includes shadows is advantageous
Encrease now the denoise level, until the ambient noise is removed but shadow detail/texture is preserved. Then leave this setting and set “Recovery Original Detail to 10-20”, leave output resolution to original size, we do upscale later! Export this work now.
Import denoised video you’ve done in Step1. Leave original scale again. Set “Proteus” and “Realtive to Auto” Fine adjust denoise if needed, then encrease sharpen until you have no artifacts, look at different scenes when do this. Don’t go to high, if you think now it’s super sharp go down often it’s to much, it has to fit for all scenes and during this steps we get a good sharpness.
Now encrease “Recover Details” and “Anti-Alias/Deblur” as a basis both on the same high value. The higher you go, the more unnatural the image can become. I usually use values from 10 to 30, The value 15 fits very often. “Set Recover Original Detail” to Zero and export the work again unscaled.
Import the video you’ve done in Step 3. Now we upscale. Choose you prefered output resolution, then choose model “Artemis…” and “Strong Halo” set “Set Recover Original Detail” to Value of 10-20. Now it’s done, export the final work.
I would not have thought of doing passes like that. Had to give it a try. I don’t think I took enough care with picking my setting. The result was not even close to what I have been getting. It might also be my source. It’s a heavy grain DVD with some very noticeable compression artifacts. Maybe it works better on even worse sources.
This method is useful if normal way gives artefacts with proteus at low settings and upscale. Some noisy worse quality vids upscale with proteus can generate this artefacts and Artemis has less of this problem.
But could be Proteus is wrong model for this sort of videos. If you get no good results you can try “Theia” It’s less invasive. The result is less shiny than you get with “Recovery Detail” and “Anti-Alias” from Proteus but this can be the better model in this case.
You get also some shiny effect and sharpness gain by denoise on point, means the veil of fog is removed, but the fine structures of textures are retained. I recommend spending time into good denoising.
That’s interesting. I’ve yet to do multiple passes in TopazAI, primarily, because I dread the thought of spending days on one video. I agree that noise can really throw a wrench into the model workings. I’ve been using Hybrid to pre-process videos (crop, denoise, PAR conversion, etc.) before feeding them to TVAI.
here’s something else I noticed. After you have used Proteus check if you can denoise the result a little bit without loosing structure. By doing this you get sharpness gain. Often it’s the case you can do this. I think it’s because proteus “burns” structures and some of the nosie into it and that’s more robust. Try it out.
I found that there is no “rule” for Iris. Although Topaz recommends it only for low-to-medium quality videos, I’ve gotten best results on a 1080i source. When I tried to use it on low-to-medium quality videos, the softening was unacceptable. My take home is that it can be very good or very destructive, depending on the source. Considering that, it doesn’t surprise me that the right pre-processing could make a source file immune to over-softening in Iris.
That said, it would have to be something really important for me to spend the time on 3 separate encoding processes. I wish that there was a combination model which simulated the use of multiple models, in one pass. For every video that I take to TVAI, I’m generally running a test segment on Proteus, on Iris and on Artemis, to see what gives the best result. Often I’m surprised.
I had the same reaction to softening when using Iris on LaserDisc sourced music videos from the 1980s. Then, I used Iris Relative to Auto increasing Recover Details to 50, Sharpen to 70, and added Grain–was really happy with the result. When Proteus doesn’t fit, I now use this approach with Iris and am able to get a 480p captures of LaserDisc up to 4K (which I could usually not do with Proteus).
I’m also going to try MayDay’s denoising approach for things I want to spend more time on, but this one-step Iris approach works well for processing a lot of content.
Thanks for posting–just tried this process and it works great! I like the approach of doing different steps because I was tending to go in circles trying to balance out all the different settings at once. I got a better end product with your 3 step method.
Thanks for feedbacks, yes in my experience, Proteus does a good job, but if you then scale up bad sources with Proteus, artifacts can quickly occur. In this case it could be better when you made the enhacements unscaled in Proteus and upscale in a next turn with another model.
Now i have to try out Iris more, as sk8townusa wrote sounds good, but I can’t say much at the moment, more later …to be continued
Every source is different and needs a unique approach. There’s no such thing as a one-size-fits-all process that works on all videos, or even a majority of videos. I’m not sure I see the benefit in tripling the time and work when, with a little experimentation, you can just find the right settings the first time?
I have found that a x4 upscale using iris (details = 100, sharpness = 100, Dehalo = 3 or 4, Antialias/deblur = -80 to -40, Recovery original = 50 to 80, and everything else at 0) as a first pass on something like a poor quality 480 source, and then take that export and resize to 720p, followed by a x2 upscale proteus run results in some phenomenal quality files.
Iris was very different to use at first but it is AMAZING and fixing some horrible quality sources. For really bad sources that I want to keep as much detail as possible, I will use something like StaxRip or Hybrid to clean it up a little (qtgmc, temporal denoise, deblock, etc. whatever fits) and then I crank up the contrast a bit. This results in Iris preserving more of the details that I want to keep.
I’m still using this 3-step method and getting great results. I’ve gone back on files I gave up on and able to get them looking really good. I’m wondering–when you export for Step 1 and Step 2 which encoder/profile do you use?
Here, I do like the Iris 2x upscale model much more than the 4x.
While 4x produces more visible „enhancement“ at first sight, it also creates funny artificial things.
Iris 2x while at first sight less spectacular does keep things very natural (faces, eyes, glasses) and doesn’t create that „comic“ look while still cleaning up artifacts very well.
I tested Iris a while now too. In my opinion Iris is Proteus with different base (auto) settings. Proteus in automode (or relative to auto) dosn’t use “Dehalo” and since V3x it uses no or low “Revert Compression” “Dehalo” breaks edges from pixel groups, it makes video smoother and this allows higher upscale. That is what Iris does. But Dehalo also makes a video less sharp. So the pros and cons somehow balance out. There may be other difference base settings Iris vs Proteus, but im sure this is one of prime difference.
You must know everytime when you upscale you get SHARPEN effect. When you scale up for example from 640p to 4k, you get more sharpen than from 480p to 1080p. More sharpness is desirable, but if it gets too much, artifacts appear and it looks really ugly.
Actually it is the sharpness which limits how high you should scale. If a video source tolerates little sharpness then it makes no sense scale it up to 4k. Sure you can take Iris because Iris does less sharpness, but the result is not a very sharp video. Or you can take Proteus and use “Dehalo” until you get no artefact when upscale. But does it makes sense when you must make a source higly “unsharp” you can make a very high upscale? I don’t think so.
Many people think more resolution brings more details. But that is quite correct. Sharpness makes details that are already present in the video (but not visible or difficult to see) more visible. That happens with the upscale, it is re-sharpened. But it is not additionally reconstructed and invented more just because you scale it much more up.
…and I wrote it already; one of the best thing you can do to gain sharpness and bring details to shine is good denoising. But I have to correct myself: A first pass of only denoising makes sense if a source is very noisy. If this is not the case (in most situations) it’s better to improve the video directly from the source. This is probably due to the fact that some information is already lost with a denoise which a model can’t use for an enhace.
Hi… It’s all very interesting. For now I am using a two step tuning method (three if we count the pre-processing)
After QTGMC (for interlaced material), I go up with contrast, and I go with one pass of Iris relative to auto at 125% scale (125% is only for activate 2x models in Topaz), with various settings that depends on the movie. After that I apply a second pass with Gaia HQ to 1920x1080. Now I want to try to apply a first pass at 100% of Proteus Denoise like Mayday suggested, because if I denoise first, probably I can go up with Deblur with Iris.
Mayday a question, what Proteus version are you using? I have seen that probably proteus v2 is generally better for upscaling, but Proteus v3 is better at 100% and at denoising.