Video Enhance AI v1.8.1

Absolutely, it is about communication. Thanks.

1 Like

hello, i have idea, recently i saw info that xbox series x and s have UWP applications running ability, what if you could make video enhance ai for xbox ? this kind of devices stand idle most of time - it could make sense to use it to calculate neural network video enhance in that time, just matter of making in app pairing with android + ios + windows/mac app so people could easly send videos to console for processing, sorry im posting this here but i cant use this forum … im autistic :c
btw it might be good idea to implement internal hardware temperature monitor to pause processing when device gets too hot

2 Likes

While my trial is still active, I should warn that Artemis-MQ 9 is giving me blinking dots artifacts that weren’t as bad in the version you had in 1.8.0 beta (before the update to the models). Those artifacts did exist, but they’re way worse now to the point I consider it unusable. Also, the denoising is way too strong in all three Artemis models. There’s no middle term for just a mildly noisy footage (you end up erasing all fine detail on those and being left with a blurry video).

1 Like

There’s been much disagreement about the new models. Everyone please post samples when you see something noteworthy (might help the developers). If copyrighted, 20 seconds would be considered by most courts to be “fair use” for a movie, considering the nature of our experiments. I’m not a lawyer, it’s just info I’ve read.

I use VideoRedo to make samples, but ffmpeg can do it too (someone who knows how, please post a usage example).

Here’s the one where I said too much CAS sharpening is making people look scary. That CAS (sharpness=1.0) can even be used without the astigmatism effect says a lot about how detailed ArtemisMQv9 is making some videos. Note only very clean detailed video benefits from settings this high. Standard def, you might use 0.2 to 0.5, or instead use vsMSharpen with strength at about 30 to 40, or use a normal sharpener (like that built into Virtualdub2).
418MB:
https://mega.nz/file/s9FhDQ4Z#U-9HP_n8w8nzcpzTQJhZrwv2XkT64Hp-pzKff2SGPKY

I’m surprised the denoising in ArtemisMQv9 is not causing any issue with clean video (IMO). I also did this video with ArtemisHQv9 and to me it looks the same as ArtemisMQv9.

I included the video done with GaiaCG 1.6.1. There’s faint raindrop artifacts and not as sharp as Artemis. Look to the green ground areas to see the artifacts.

If you don’t know CAS, Contrast Adaptive Sharpening, it’s one of the best algorithms I’ve ever used (IMO), comparable to vsMSharpen but works on more than just edges.


http://avisynth.nl/index.php/CAS
http://avisynth.nl/index.php/VsMSharpen

1 Like

All right.This sorts it out :slight_smile: I really didn’t pay any attention if it is a public release or release candidate.I just read the release notes that it states that we can keep our models and downloaded from the link given.Also as I mentioned before, it got installed as a release version,so I didn’t feel any difference there.Another thing is I am not in the facebook group or among those beta/alpha testers for a long time.So I don’t have any clue that it was still in decision making stage.

Anyway,let’s make it straight. If you checked what’s been written people tends to keep their models with backing up to a different location,they don’t want to re-download the same models for various reasons and most importantly efficiency/time. There is no reason to hold “keep the models” feature back, since otherwise we are forced to do the workaround.

Make it public on the next releases and it is all good :innocent:

Edit : Also please consider making small patches instead of making the whole pack to be installed between versions if possible.

1 Like

By the way, I think it’s a frame rate bug because on a 25 fps TV series, it works fine. But on the movie in question which is 23.976 fps, in Topaz Video Enhance it says 23.98 fps. So I conclude that for the moment Video Enhance doesn’t handle 3 digits after the decimal point.

1 Like

Another point, I also noticed that the rendering time was the same when I went from 720 x 576 to 1920 x 1080 or when I put one at a scale of 200% (2048 x 1152). Rendering speed for both: 0.16 seconds per frame.

1 Like

Artemis LQ v9 is quite amazing, but, it does something remove details that it “thinks” are noise, which are not noise.

I would like to see Artemis MQ and HQ be like LQ, but not be as aggressive at removing noise. Right now, the Artemis MQ and HQ models seem to be completely different than the LQ model, and are kind of blurry.

2 Likes

a little blurred if you go from 720 x 576 to 1920 x 1080 for example. But if you go from 1280 x 720 to 1920 x 1080, you will get a really sharp, grain-free and very precise image.

1 Like

Little trick to get back some sharpness, I import my image sequence in Premiere Pro and add some sharpness.
It’s quite effective.

1 Like

Artemis v9 requires very clean source material, but even LQ is crushing detail way more than GaiaCG from several moths ago. I re did a few files that I had done with GaiaCG back in May and it’s startling how poor v9 is in comparison. I can only assume that most advocates are using very clean HD sources for upscaling and , well, what doesn’t doesn’t look good when using a HD source?? But for my purposes SD to HD, v9 is a wash.

A constantly evolving program is never going to satisfy all, which is why previous model versions need to be accessible, because each one will work in some way for some. The way it currently is, means a model that you find works with a certain type of source, then gets replaced by a new model and you lose usability. Trying to find a one fix for all is never going to work.

3 Likes

Yes, GHQ is a great preservation model. It gives a much more natural look to upscaled material. Av9 is like an industrial polisher. I’m just pissed because I can’t use that model anymore, making the program less useful.

That should be 1440x1152? In any case, for your 1920x1080 VEAI would have used 200% and then done a simple resize, with the latter operation being too fast to notice a speed difference. You may not have noticed other discussions which tell us that VEAI only does 100/200/400 and all other upscales are resizes from one of those.

Did you ever look at the frame rate of the one that was short after processing? I think you will find that it is 25fps, and you would find that if you upscale a 30 fps movie it will wind up long.

The bug occurs (apparently) if you ever input an image stream to process, although there may be other ways. When you do, VEAI sets your framerate permanently to 25 for all future renders. Only manually editing the registry will fix this.

1 Like

Should I apply sharpening to my 1080p timeline while editing and exporting before upscaling? Or skip the sharpening, upscale and then look if sharpening is necessary on the upscaled file?

Can someone make a step by step guide how to avoid exposure shift with hybrid and VEAI? If I upscale 1080p master > UHD the exposure shifts quite significantly. I tried using Avisynth but the final result is very low in size, can’t play it with VLC or in PP. What do I need to make Avisynth filters properly working in hybrid?

1 Like

Yep of course… I know its slow hardware, we just have a need to use this smaller slower laptop for certain location reasons, however it did work on the v1.7.1 at an acceptable but slow speed so we were asking why the v1.8.1 should crash when attempting to use the same GPU and hardware as even if the GPU is slow it would be more graceful to have error messages and not to crash or just to run slow but usable like in 1.7.1… :slight_smile:

1 Like

How do I use Avisynth properly when selecting in hybrid the exported file size is low in size. VLC doesn’t play the files.

1 Like

I see.

1 Like

Just to add to that:
if you go up until 119%, the app will use 100% and lanczos to resize to the desired res
From 120 to 239% it will use 200% and lanczos to resize to the desired res
from 240 and up, it will use 400% and lanczos to resize to the desired res

1 Like

Hi there, when it is ready and thoroughly tested it will be released. We cannot guarantee an exact timeline for any feature in BETA, but we do understand that you are very eager to have it.

It is still being developed and is not on hold. We appreciate the feedback. Thanks.

1 Like

Thank you for reporting in, we have logged this issue.