Video Enhance AI v2.3.0

Topaz Video AI Releases
102

Friend lee.keels the idea here is not to fly blind, I believe and many here want to know what they are working with, and I’m sure many here from both Topaz support and beta testers know exactly what each term means to me posted, and that most beta testers are not doing blind flight, the trial and error type to verify any problems of the models presented so far by VEAI.

Here is a small example of the video Yanni - Tribute from 1999, what are these effects occurred with Proteus using Model Parameters AUTO.

Here is the 17-second generated video:

Att.

Druid®.

103

It looks like it’s picking up some leftover interlace artifacts. If that’s the case, it may not be the fault of Proteus.

104

Did you see the Proteus video here Video Enhance AI v2.3: AI-driven frame rate conversion, realistic slow motion, and so much more!
It might help a little.

In addition, revert compression (reduce blockiness when video is too compressed, for example to make a long video fit on the DVD media), sharpen (increase sharpness if image looks too soft), recover fine texture and detail (for example, if you want to see every hair and the texture of skin), ISO noise (when the image is grainy), dehalo (reduce the outline around edges of objects), antialias (reduce the ‘stair-step’ edges of computer graphics by moving the slider to the left of centre, deblur (if the image is slightly out of focus, move the slider to the right of centre to increase focus.

2 Likes
105

Even if they were just named Model A, Model B, C, D, etc., my advice would be the same. You won’t know what it does for you until you try it, regardless of what the devs or other people write about it.

106

Task Manager does not reflect CPU and GPU accurately. Never has.

107

I’m afraid your RIP might not have been properly de-interlaced by the looks for it, causing Proteus to “reveal” those flaws rather than be the cause of them appearing.

I couldn’t agree more mate…

In every project of mine, I have to go through the painstakingly slow and time consuming process of trying all the models I deem fit for the source (including previous iterations of the artemis models) and compare them side-by-side (using snippets from different scenes) in order to finally make a decision.

And that’s just good quality HD source; For SD’s I usually end up pulling all my hair out by the end of it, since it now involves me using various 3rd party filters and upscaling algos (and compare the different results (ugh!)) to get it into a decent shape before feeding it to VEAI.

This stuff is no joke lads - You can’t attain video enhancing nirvana by just the press a button. :persevere:

1 Like
108

Edited to clarify what I meant:

Yes, time consuming to compare all the models. I meant I could do a quick run of each of the sliders on Proteus, just to find out what they do within 15 min or so.

110

Mine shows 5% GPU usage on average. Task manager is not THAT inaccurate.

111

I did misread what you wrote…I read it as “couldn’t disagree more”…LOL. And I agree with you about it being a long process testing a model and comparing them. I was specifically referring to testing the individual sliders on Proteus only. That should be a very quick process to just see what they do. That’s all I meant.

1 Like
112

I know you wanted it done by AI (me too!), but at the moment you can do EXACTLY that with a couple of AviSynth scripts, if you use it. In case you’re interested, let me know and I’ll post the scripts here.

113

Apple CoreML (fp16)
https://pastebin.com/RZN4ZxSB
52.1 GB

Apple CoremL (fp32)
https://pastebin.com/dq6eXfNp
95.0 GB

Intel OpenVINO (fp16)
https://pastebin.com/zeifRc7E
50.6 GB

Intel OpenVINO (fp32) (Note: missing many models)
https://pastebin.com/Dzqduxuy
24.5 GB

Open Neural Network Exchange (fp16)
https://pastebin.com/rincBudr
49.5 GB

Open Neural Network Exchange (fp32)
https://pastebin.com/fF1MjsMF
99.6 GB

Missing (offline) models
https://pastebin.com/wkCTS3Mf
(Note that these models will likely never be uploaded by Topaz.)

To get different model versions to appear in VEAI’s model select list, open the JSON in a text editor and change the version number on line 3. So for example, to get Artemis LQ v8 to show up on Windows, open C:\Users\[USERNAME]\AppData\Roaming\Topaz Labs LLC\Video Enhance AI\models\artemis-lq-12.json in Notepad, and change line 3 "version": "12", to "version": "8",, then save the file as C:\Users\[USERNAME]\AppData\Roaming\Topaz Labs LLC\Video Enhance AI\models\artemis-lq-8.json.

You don’t need to download all the models for the above JSON edit to work. Video Enhance AI will only download the models it needs based on your selection. You can use this technique to determine what models your system uses.

GPU on Windows = Open Neural Network Exchange
Software render / No acceleration (EXTREMELY slow) = Open Neural Network Exchange
Supported CPUs = Intel OpenVINO (should also work on AMD)
Apple = Apple CoreML

Most systems will likely use fp16.

One exception… the new “automatic” Proteus model proteus-auto-1.json model only uses OpenVINO fp32; it’s one file: prap-v1-fp32-ov.tz. (It’s installed by default.)

The installer installs Open Neural Network Exchange (fp16) 288x288 models, for current models only, and prap-v1-fp32-ov.tz (+ prap-v1-fp16-ov.tz, which is not used as of yet).

3 Likes
114

thank you for the answer.
My config is:

Intel core I9 9900K of 8th gen
32 Go of ram
RTX 2080 with 8 go of ram

115

But there is no way to have another link than the pastebin? Because it might take some time to do one by one. Moreover with internet download manager, it converts all to zip file. And there also I will take time to put the tz extension.

116

I heavily recommend JDownloader as an alternative to IDM especially for that kind of purpose, since all you need to do is to select and copy the links to the clipboard and it will automatically add them to the link collector (download queue).

Windows ad-free links: 32bit - 64-bit

1 Like
117

I’m noticing a few bugs with the latest release.

  1. Open the program and drop a video to import
  2. Click “Clear All” and click to confirm
  3. Drop the same video into the window a 2nd time
  4. If you then hit “Preview” or “Start Processing” nothing happens and it gets endlessly stuck on “Estimating ETA…” You can hit Stop and then Close leave the preview screen but if you did Start Processing you can’t Cancel out and your only choice is to terminate the program with task manager.
  5. I’ve tried this with multiple videos of different formats and it keeps repeating.

You also changed the behavior from the previous version and now “Select All” is checkmarked by default when you first load a video. This really sucks because I have to uncheck it if I load a 2nd video or else whatever model I pick for the 2nd video overwrites whatever settings I had for the first. Also if I drop in a 3rd or 4th video they’re all checkmarked as part of the same group so I have to uncheck them all one by one as I load them. This is super tedious so please change it back!

118

thanks

1 Like
120

I was away this week and couldn’t try the final Proteus at the last beta but now it’s already in the official released version and wow it can bring amazing results with the right settings.
Less on the dehalo alone with VHS input where Artemis strong dehalo is still better.
But I wished Antialias and Deblur could be seperate settings because we can have an aliased video that also need to be deblured.

1 Like
121

Is ALQ13 missing from 2.3.0?

122

For some reason, Artemis LQ v13, Artemis LQ Dehalo v2, Artemis MQ v13, Artemis MQ Dehalo v2, and Artemis HQ v12 were in the last beta (for 2.2.0), then not included in the final release. The same goes for Dione TD v4, Dione TD Dehalo v2, Dione TV v4, Dione TV Dehalo v2, and Dione DV v3. Since they’re not included in the 2.3.0 release either, I’m left wondering if they’re going to get any more updates.

The model files are available for download, so you can create JSON’s for these by changing the version numbers and saving them as new files. Personally I really like Dione TD v4, it’s my new favorite model, and I’m disappointed that it was left out of the official release.

1 Like
123

So how do I sync an audio track to a video clip that I increased the fps with Chronos? My original clip was 6:40 and 25fps. After using Chronos it’s now 60fps but 6:58 long, 18 seconds longer than my original audio.