Eventhough I worked in more places in the US even compared It is still my great wish to travel Utah and yes the mountain of gems is on the list. You are a priviliged person.
I never use rhea but I had abba big weekend VH1 recordings in 2009 the surce videoclips from the 70’s and 80’s were not good this time I always use proteus V4
I always first let it convert to HD in video proc pro to 16"9 to HD auto copy let it deinterlace h 265 gop 30
I use the Rheia UHD settings Normal this will only do minor enhancement the normal this is fast enough and Yes it is to sharp at times to white teeth but certian clips do incredibly well dependent on the age 74 is not good but most clips are close ups and look ok hence since there must be a hugh amount of different tapes the quality is not perfect anything from 76 is ok probably they started to use better equipment
Tried again with the Video Proc Source but now Iris lq v2 UHD Normal setting, but adjusted the dynamic setting improve detail 49 and set the Alias Deblur to 55 left the remainder 0 that is in the middle since the video clipse are all mainly faces I have to say this resulted in perfect results natural detailed cleanup but I would only use it for videoclips vhs
Yes, Iris LQ v2 is well-suited for enhancing video clips from the 1980s. It’s designed to improve details in low-resolution footage, making it a good choice for older videos2. Users have reported positive results when using Iris LQ v2 for upscaling and enhancing VHS tapes and other low-quality video sources from that era.
X factor UK 2008 etc 100 Protheus
homevideo digitized by hardware or other legacy stuff nero this is likely out of sync and I have to use progressive to interlace iris lq v2 which is made for correcting the artificial stuff if you use progressive
Proteus Manual or Auto/Relative ?
They are both different Models.
Manual works mostly Spatial (one Frame, or at best 2 Frames Temporal) and Auto/Relative works with 2+ Frames Temporal.
Temporal is better to keep Structures in Shape (that’s why plain and simple Artemis can keep Shadow Spots like Fatpads^^ on Legs better) but will eat up Pores & Pimples if the Motion is fast and the Pore & Pimple Size match the Size from the Noise around.
Spatial will keep anything that is under/over (it depends on the Filter) a certain Treshold.
The best is if something works Temporal (analysis over 2>= Frames) but with Strenght Values for the Spatial (1 Frame) Part.
If People in the background are unsharp, just take it 
it’s natural that something more far away has less detail. That’s Physics & how our eye work, so it should have less detail (our Brain say this to us. It’s trained for that)
There’s a reason that the Main Role Cast is in the foreground, and the rest is in the Background. Main Role & what they do is important, rest is not important.
1 Like
Models used in earlier versions? If like me one has downgraded to an earlier version, are the models used the current ones or are they also rolled back to what they were with the earlier version? I hope that makes sense
Try Rhea at 1x size / without upscale:
Very good cleanup of old messy vids and also some details gained but (mostly) without those annoying things, like monster faces, creepy eyes, repeating low res textures,…
See here for examples:
2 Likes
As far as i know, newer Versions will download Versions that work with the newer Version, but still call them something like “Artemis - Medium V13”
If it changes the Output when the same Settings are applied can only be checked with one way.
Take old Version (let’s say 5.4) create File with Artemis Medium V13, save as Uncompressed.
Do exactly the same with 5.3
Then compare the Output from both Frame by Frame.
Or
Delete any Meta Data,
and make a SHA2 or Blake 3 Checksum.
If the Checksum from both are the same, then both outputs are absolute identical Files.
When i see that Topaz say that a newer Version is now faster, and no Mention that Model Output is different, then i always ask me. “At what cost ? Did they checked if the Output is 1:1 Identical ?”
I see people making Benchmarks, comparing different Versions, and get exited if some Version got 1-2fps more ^^
It should not mater if some Version is 1-2fps Faster, it should mater if it’s faster with identical Output !
Never Delete your Original Files ! Always keep them, and allways ad the exact Topaz Version to the Filename from your Project. Especialy if you want to Share your Project File with other, so that they have the exact same Output Result, when they process the same Sourcefile.
1 Like
Thank you for the clarification.
I use Auto/Relative most of the time, because there’s so many scene changes inside of the source that it’s impossible to make an educated guess about the best parameters at all.
1 Like
Yeah i know what you mean it’s way faster with that to finish the Project and will still often result in a Output that satisfy. (but there’s always one param that i screw realy down and that’s Dehaloing. It’s a Pore Killer and can transform thin and very short hair^^ to little dots or create strange looking very thin lines that just look unrealistic)
It just gets problematic if the Output has ugly AI artefacts in the Face, or when the detail grade alone is compared with Optimal Proteus Manual Settings for a Scene.
With Manual you have so much more work, and need to work with small parts, but can achive brutaly good (and better as Auto) results for that specific part.
If the Part is not small you will sooner as later see Detail on Skin that come and goes (especialy if the source has Blocking Artefacts & Noise that is hard to seperate from real Detail) the frames Pump.
Pumping Frames: Detail on leg^^ is there for 2 Frames, then gone for few frames, 2 Frames there, then gone etc etc
With Auto, Pumping Frames (and especialy Aliasing on Hair Motion) risk is kept much lower.
In the End it’s a question about how much Time somebody is ready to spend.
If you use Auto for a 90 Min Movie, and it took 10 Hours, and later you have to split up the Output, because there are 1-5 Parts with screwed up Skin or Faces, then you lose a Ton of Time, and got to do the same thing that you should have done in the first place (smaller Parts)
If you used Video AI enough, you get a feeling what sould be used.
Longer Part need Artemis X because of Noise Size, Noise Distribution, Noise Speed, Blocking Size, Blocking Speed, Blocking Distribution, Face Distance, Speed of People Motion etc
Next shorter Part with just a few Frames need way stronger Settings because…
Next longer Part need Artemis Y because … etc
That’s why i always say that Topaz Video AI itself should have a real proper Joblist so that user doesnt need to administrate all the Cut Out’s externaly (and there Filenames) by his self.
Writing something that feed’s Video AI with Cut’s is no problem, but it makes only sense with Program Exit Codes (with that the Tool that started Video AI in CLI mode, knows if anything was Ok, and if not, what was the Problem) and a “100% Windowless/hidden CLI Mode”
Another thing to consider. FFMpeg feeds Video AI, and creates the Final Output from Video AI, so Coder must hope that FFMpeg get detailed enough Exit Codes (if something went wrong) and pipe out the Exit Code from Video AI.
That’s why i wrote my Beta Cut Out Tool (it send’s Cut Out’s to VDub2 with AviSynth Commands. Waits until VDub2 is finished, and anything was Ok, then it executes the next Cut etc etc and when anything is done, it joins all the Cuts to a Final Output File) just for AviSynth & VirtualDub2.
Because VDub2 has a variety of proper exitcodes.
2 Likes
My license expires in a few days. What is this release of TopazVideoAIBeta-6.0.0.3.b? How is it different from 6.0.0?
As far as i heard, it has les bugs xD
There is already a newer beta version 6.0.1.0b, which works better than the final 6.0.1 version.
https://downloads.topazlabs.com/deploy/TopazVideoAIBeta/6.0.1.0.b/TopazVideoAIBeta-6.0.1.0.b.msi
If you want to use a final version, which is not beta, the last one that works well is version 5.3.6
https://downloads.topazlabs.com/deploy/TopazVideoAI/5.3.6/TopazVideoAI-5.3.6.msi
2 Likes
What is really good about Topaz!
- Easy file conversion.
Just select parameters and codecs and save without processing.
I needed to convert a Blu-ray file MPEG_TS VC1 to MPEG_4 AVC, or FFV1 for further processing in Davinci. Other Avidemux or Shuter Encoder programs have shown that they will convert 8-9 hours with a heavy load on the processor and a temperature of 80+. Topaz did it in an hour at a processor temperature of 69-70. - Improving quality 1080p and 4k videos to even greater quality. Rhea or Proteus is best, but if you adjust to no more than 5-7%. If you set more, then nothing will save you from distorting the original. That is, Topaz copes well with a little “polishing” of an already high-quality video.
- Almost the only, main, unique function of Topaz is frame interpolation. I’ve tested a lot of apps. None of them can make such a smooth picture at 60 or 120 fps as Topaz. But only Aion or Apollo. Cronos is absolutely unusable!
Now about the bad in Topaz!
This is not about meaningless jewelry, like a preview or the appearance of windows.
It seems to me that there are no AI models at all. There are simply ffpeg presets named by various romantic or mythological names.
Why do I think so. I’ve tested many options for medium-quality video snippets on different models. Whatever the name of the model, there were the same distortions, the same artifacts in the same places. More or less manifestation.
Whatever I did, no matter what models I choose, all distortions and artifacts were in place. Therefore, I chose the only way out, to improve videos only in models with manual settings.
What are the changes since 6.0.0?
The preview is still blank (black). I can hear the audio, though. Not sure why it doesn’t work, I may have to roll back to v5.3.6, again. Very disappointed that this bug hasn’t been fixed.
2 Likes
Your Right there is not AI, just well trained Models
Real AI would understand where Persons are, and not add Grain on the white wall 
But it is still way above anything that ffmpeg can achive if the Source is Garbage.
If a trainied Model fits, then those Models can do Deblocking on a Level that you can absolute forget with ffmpeg. You can achive them if your are a AviSynth Guru and got a looooot of time.
If Topaz stuff is used for a allready good Source, then it’s no contest at all, and the Problem is the User (with his to big screen^^) and his “k” Horniness
If Topaz is used for realy bad Source,s then it shines.
(Look at my Second Post, that’s what i call Bad Source and Blocking)
Why are you change stuff to 120 fps ?
1 Like
Thank you dearly for your detailed explanation.
I’ve got a ton to learn
2 Likes
Alternative to Topaz there might be other software,but it is very expensive as a subscription, but from what they say it must be very professional.
Thank you, but why should anyone think that one would be any better than TVAI? It’s just a list of the same claims made by Topaz, by a different company, one that even needs support via Patreon!
Until AI is able to identify and track individual people, faces and objects, pick the best quality most realistic frames and use that level of quality for ALL frames where the same person/face/object appears, AI video enhancement - like Stable Diffusion etc., will remain where it is now: in its infancy. And it would need to do that on consumer hardware which is an even bigger ask.
That would be real AI rather than the guesswork that it does now when things are partly obscured, go blurry or simply lack detail (as with small faces on very old footage). Yes we’ll get there in the end but I suspect that is still years away.
Meanwhile, for enhancing and repairing very old footage with lots of bad frames automatically (minimal intervention), which is my use case, I think TVAI in its versions before 5.0 and maybe back to 3.x, is the best we are going to get. Version 6? It’s no better for my uses, no better at all.
3 Likes
I am stuck on version 4.2.2
1 Like
That’s completely made up. No one from Topaz has ever said that. They only ever said it was made to be sort of a mix of Proteus and Iris.
As far as this statement goes:
There must be a threshold of how bad the source has to be for it to look better. I tried a mid 2000s DVD, and even at 1X scale, it made the skin look more plastic and baked some of the noise into spots that are impossible to clean out later.
2 Likes
