Topaz Video AI v3.2.0

I understand that it takes time to see how it works. But I have only one request. Just make the balance darker so that the video is not a little lighter on dark frames, so that it is like on the previous firmware. And remove excess noise (gain) when 1080p and 4K video is processed. Noise can be useful when processing low-quality video to mask processing distortion.
P.S. Apollo stopped increasing the FPS on video in 4K. Gives an error. On 1080p works!

1 Like

I got Apollo to interpolate to 120 fps with a video upscaled to 3840x1600 (the same res it shows yours at).

You could try using the option to open the log folder and when youā€™ve finished your current work in the app, close it so it closes the log and open the last log and see if it shows what the errors were that caused the ā€œunknown errorā€ to be shown in it. You could also click on the 3 dots next to ā€œunknown errorā€ to see if it shows a shortened bit of info about what the problem was there.

Uninstalled the program. Reinstalled. Everything works!

3 Likes

I have a really simple request. Please make adding an icon to the Windows desktop an option in your installers. Not all of us want an icon automatically added to the desktop with every update. It is especially annoying that the way Video AI icon is added requires logging in as an administrator to get rid of it.

Thank you.

Does anyone know how to turn off that godforsaken ā€œCLOSE ALL SOURCESā€ warning when youā€™re getting rid of a clip???. The extra clicking is driving me nuts.

In my opinion, itā€™s the way your machine is configured because on mine, the icon is duplicated with each update, I donā€™t need administrator permission to remove it. given I am the administrator of my machine. Unless you are on a MacOS machine?
Me, Iā€™m on Windows.

  1. How do you start a totally new topic on this Topaz Video AI Link?

    The only way I know is to reply to the first post.

  2. What is the progress on the faces being distorted?

  3. What models are best to used & any suggestions on processing with lease facial distortion?

Thanks!

Mike

You could go to:

and select one of the options in there - some allow new topics created by the user. eg. you could select ā€œGeneralā€ or ā€œIdeasā€ then select ā€œNew Topicā€.

1 Like

When are you gonna fix the performance? This isnā€™t normal:


It should be a constant GPU usageā€¦ And above all, it should use as much VRAM as possible, not 5GB. Itā€™s not taking advantage of the graphics card at all.

1 Like

As I said in my request, I am on Windows. I have decades of experience with security on Windows, so I do not do my day-to-day work as an administrator, only as a standard user.

It is a simple request, one that all other software developers do. Give people the option of not having an icon added to their desktop with every update. Thatā€™s not too much to ask.

A much-needed hint regarding TopaazAI.
I really love the video being as clear and very smooth as possible, so I always put 120fps. I just donā€™t perceive cinematic abomination 24\25\30fps.
I now only process high quality music videos where there is a lot of movement, dancing, slow motion, different transitions.
I still havenā€™t used Motion Deblur when converting to 4k120fps. Will this video feature improve in my case?
The second question does not concern TopazAI, but is general.
For example, if the video is 24 or 30 fps, then the increase to 120fps has multiplicity. 24x6=120. Or 30x4=120. And what to do with 25fps? It would be logical to do 100fps (25x4=100). But the TV has 120Hz, or 60Hz.
So, for smoother playback on the TV, is it better to choose 120 or 60 values for all initial fps? Should the initial video do 25fps or 50 fps?

Iā€™m sure this has been mentioned to you before, but you can input directly into the frame rate box any frame rate you like, including 100 fps. I just did that with a 25 fps clip, also I did 95 fps (and couldnā€™t tell the difference but OK it was Adele and sheā€™s not exactly a fast mover lol).

As they have a 50 fps option in the dropdown, I suspect that eventually, theyā€™ll add one for 100 fps too but it will only save a second or two per clip. It has to be better to choose 50 fps not 60, or 100 not 120 fps - but most modern TVs can handle either automatically (if their refresh rate is high enough).

1 Like

every time i process a video to 60fps with apollo, the end result has this stripe at the bottom with no color.

image

It is using the GPU, you just donā€™t know what youā€™re talking about.

This is using TVAI 3.2.0

Screenshot_20230408_220710

1 Like

Incredibly cool! For some reason, I did not pay attention that you can enter the value manuallyā€¦
Hold on, guys! Now we will do 240fps, 240Mbs. I donā€™t need compromises. Posterity, if there is no nuclear war, will thank me that I kept all sorts of shitty in this capacity.
But please, if possible, adjust the color and noise a little by next Tuesday. I would be very grateful!

1 Like

Does it seem normal to you that a 4090 is using 5GB of VRAM and is at 60% usage with constant ups and downs at 1-3fps when scaling to 4k? Have you ever used other AI tools for video processing? It doesnā€™t look like it, because you normalise the lousy performance here.

There is still a strong white halo in the low-resolution video. I tried the strong halo filter or AI model, but it removes the dark halo only, not the white halo. And honestly, the result is ā€œOKā€ for the dark halo, but not really efficient because it adds a strange soft glow effect, with low bitrate and low-resolution videos at least, and needs to be improved. And yes, Video Enhance AI also needs a white halo filter or AI model. It would be great!

Dear developers, would it be possible to include the script to download all models at once (as a separate file) with each release of TVAI, whether alpha, beta or final version? The last script posted by Suraj no longer works.

This would would be a welcome workaround while we wait for a models manager to be built-in at some point in the (not too?) distant future.

Here is the same task manager but this time im running a pytorch SwinIR model on the same input video. I think Topaz have done a pretty good job on the performance side of things.

Screenshot_20230408_224237