Video Enhance AI v.1.2.3 - Gaia-HQ model improvement (hotfix)

I have waiting to buy this software for months already.
I have only one most important problem stop me from buying.
When will it support AMD Graphic Card ? ?
The processing speed using CPU is extremely slow and totally unusable.
I have a Ryren 7’s 8 cores CPU overclocked with liquid cooling but the processing speed is still far too slow to be usable.
It need to support AMD GPU to speed up processing.
All other Topaz AI such as Gigapixel, Denoise AI, Mask AI, Sharpen AI, etc. support AMD GPU why only the most processing intensive VEAI don’t support ?

2 Likes

According to tech support they are actually working to get AMD GPU support. But it’s not as easy as they had hoped. I’m waiting also.

Why is the out-of-sync audio still not getting fixed?

2 Likes

If you start a gpu intense game or other application at the same time, VE AI crashes for example. I have a 2070 super gaming x card too btw.

1 Like

That has pretty much been a known issue for awhile. Must be hard to fix. For me it means I can’t do anything with Gigapixel while VEAI is operating, and I also do a lot of 3D rendering that has to be on hold when working on a video. Hate to have to buy another machine!

Maybe you can use Gigapixel via CPU while Video Enhance is working with the GPU. :grinning:

Definitely worth a grin! Or maybe a groan…

Or get a 2nd GPU :smiley:

lol… I had RX 5700 XT and bought a RTX 2800 Ti just for this!

1 Like

Any reason why processing time per frame would drop from .3s/frame to .5s or .8s/frame when upscaling a 60p video vs 30p?

The new versions since 1.2.2 take forever to initialize the AI engine, more than 60 seconds vs 10 seconds in the previous versions. What’s going on ?
This is why I never used v 1.2.2.
About GAIA-HQ improvement, I have tested one sample I made with v 1.2.1 and on some areas I prefer v123 and v121 on some others. Really depends on the subjects.

Output to PNG files and then use FFmpeg to make a lossless video from the PNG files

ffmpeg -framerate 23.976 -i %06d.png -c:v h264_nvenc -preset lossless output.mkv

2 Likes

I am not sure if you are aware of this, but AV1 is currently very very far from being introduced to mainstream due to its extremely slow encoding speed…

This is a weird combination

You are concerned with quality (lossless) and yet you want to use nvenc? Doesn’t make any sense.

1 Like

Bummer. Maybe in the future it can be optimized better.

1:
Anyone know what is Uscaling mean? Maybe Upscaling not Uscaling.(here it is the picture : i.ibb. co/mGZNR58/1.jpg) And can you make a new users guide for the new AI models.
When i use the compuer generated content AI (GAIA-CG)? What does it mean “computer generated” It means cgi movie content or what? I don’t know.

2:
Guys have you seen upscaled content from c & c game? They upscaled a 320x176 video and it looks amazing. How can they do that? I tried the same with topaz but it doesn’t look as the same.
Here it is the original:
youtube. com/watch?v=FHUer1WKUgA

Upscaled:
youtube. com/watch?v=1uJXURjdIV4

I can’t post links idk why, just remove the space front of “co” at the picture and “com” at the youtube video.

You are correct. I tried to use various models with Video Enhance AI on the same video from youtube and the results are not as good. But we don’t know what they used as source. You tube already compresses the file heavily and we don’t know if that was the original.

However the file is 15fps and its not very long, just short of 500 frames total. So that is something you might be able to deal with in Gigapixel AI and simply break the original video in image sequence, like PNG. And than in Gigapixel batch upscale the sequance and recompile it back again in some other program into video.

Here is what I got using Gigapixel AI. I think it looks even better than what they managed to get.

P.S.
The computer generated mode in Video Enhance AI (GAIA-CG), can be very usful to remove heavily compressed footage and noise reduction, and you can add some film grain on top if you feel its too smooth, but I was able to salvage previously unsalvageable footage that was not CGI, it was just badly compressed with lot of compression artifacts. The AI (GAIA-CG) simply reconstructed everything.

Just practice and experiment. Its not all working as intended and in some cases its better and it other cases worse.

Cheers!

1 Like

@itunes663
Hey thanx for the answer.
They used the ps1 version videos instead of pc, it’s better quality. I can give you the raw file because yt compressing it. I tried also gigapixel ai with auto settings to upscale image sequence. But it looked like a cartoon movie. :-
link: https://drive.google.com/file/d/1rG6jvWWf2O-Mt5GGdhp-CqtSlUJp9kkb/view?usp=sharing
remastered link:
https://drive.google.com/file/d/1oaoZzy4XE8iITWhU6stXzb533bw60nfz/view?usp=sharing

1 Like

Oh, yeah that looks a lot better coming from a better source. Thanks for sharing.

They might have used another AI model from another company and some other methods.

Coming from very low res, close up shots including people mostly look OK. It would be a different story on a wide shot to reconstruct the face properly.