Topaz Video AI 5.1.4

Have you actually tried MoMo before making such bold statements?

I downloaded “Video Proc Pro,” tested it briefly, and then deleted it. It’s garbage - unusable for over 90% of video files. Topaz Video AI is light years ahead.

Video Proc Pro also has an image upscaling variant (“AiArty Image Enhancer”), which is slightly better. It can be useful in certain situations. They are currently running a giveaway for a one-year license with promised updates. Aiarty giveaway (You need to click on “license giveaway” and use the serial number). It might be helpful for someone.

However, I must emphasize that both Gigapixel AI and Photo AI are far superior to AiArty Image Enhancer. :slight_smile: Ultimately, everyone will use what works best for them.

VideoProc Ai is very good but not versatile at all. some vids are awesome, some other are pure shit.
If you matched both kinds of scenario in same movie/video then it’s somehow useless.

You’ll have to cut/copy rework scene by scene … See you in next century

When exporting in H.265 and selecting “Copy” to copy the audio of my original videos over, it doesn’t… The upscaled videos have their entire audio removed. This only happens when processing videos in bulk. Has to be a bug

No one is more enthousiatic about topaz ai as i follow my beta replies…at least the developers understand what I see most of the communitymembers not this is not an offense it is what it is

H265 Main10 Constant bitrate still doesn’t work.
Set (for example) 24mbps and it outputs 1600kbps, way short of target… High does not have this issue but is limited to 8bits.

I’ve also noticed that metadata is incomplete with Main10 (compared to high), often losing some or even all (it can vary quite a bit depending on source) of what was in the input.

Writing library (AMF for AMD and NVENC for Nvidia) is also borked, with topaz ai seemingly not changing the parameters when you change GPU selection in settings.
In my case, I swapped from AMD GPU to Nvidia in topaz settings without restarting the program and my output was tagged with HEVC_AMF rather than the correct HEVC_NVENC. My Nvidia GPU was being utilized for the render, so that’s not the issue.

I’m assuming (haven’t tried) that changing GPU selection currently requires a restart of the program to take effect properly, so it should have a prompt to “save and restart” when changing that in settings the same way that disabling “anonymous data collection” does. or just fix the parameters to match the GPU selection as other software does. There is no dropdown to select “AMD” or “NVIDIA” in both “Main10” or “High” like there is with the “Main” profile for H265 either.

Example of H265 High vs H265 Main10 on my Nvidia card, both with Constant 24mbps set. other settings identical. Output using AMD is equivalent, but with negligible bitrate differences.

I just tried this in demo mode and it was not a good experience.

The preview is not useable. When I select 100% preview for an upscale, I get a random portion of the upscaled video and only that random portion. I can’t grab and move the video to see various parts of the video. It just takes that portion of the video, crops everything else off, and then allows me to move that rectangle around. I figured out how to “open in external player” but the windows media player doesn’t do frame-by-frame. With a properly working preview, I could stay in the app, move the video around so that I can scrutinize whatever portion of the video I wish, and go frame-by-frame. But the preview is broken, so I can’t do any of that.

I also discovered that moving the crop (the actual crop feature) in any way dumps all the enhancement settings and requires me to start over. So if I get the enhancement settings where I want them but decide to move the crop window, I have to start over.

This is a bad experience.

I knew there were some bugs but thought I may renew my license if there’s a good sale. However, I don’t want to spend any money to experience that frustration.

I like the idea of the model manager in 4.x.x, but the model manager doesn’t allow me to download all the models without an error, so it still needs to download some models when I run them.

I have rolled back to 3.5.4.

1 Like

I have also noticed strange things happen while using H.265 Main10 that weren’t there before. I get AWFUL banding and artifacts now.

Agreed… My Main10 output has banding and artifacts as well.
Looks like the encoder is eating most of the grain (that I want to keep) due to bitrate constraints which isn’t helping that problem. Again, not an issue with “High” profile so something is definitely off :confused:

Edit : I’ve done some testing in command line and managed to get H265 main10 constant quality working in TVAI. Export command from TVAI and replace "-b:v" "24M" with "-rc" "cbr" "-b:v" "24M" before pasting it back into command line (CTRL+T within TVAI window) and constant bitrate is now working properly. So yes, this needs to be fixed from Topaz’s end, but there’s a workaround for now.

Orr you could use -cq which is better than constant bitrate. Swapping out "-b:v" "24M" with "-rc" "vbr" "-cq" "12" "-qmin" "12" "-qmax" "12" "-b:v" "0k", treating the cq value similar to how you’d treat CRF (which is not available in NVENC).


One thing I’d like to know is what architecture(s) Topaz chose to build their models, or if they developed their own thing. I would really like to train my own model, but the open-source codes for video super-resolution are way ahead of my capabilities. I can train my own image super-resolution model, but isr models are not temporally aware, just spatial. I think we still have to mature the super-res field a lot, and it would be awesome to have something open-source that is not so complicated.

Y’know this makes sense why my exported main 10 videos suddenly were comparatively tiny in size

the link is not working
where can I download it?

Agree, being a software engineer and now manager of development, I sadly see some of my team create bugs on purpose thinking it’s job security. This team “activity” triggered me to start code reviews which are intended as a learning process to help not hinder. Unfortunately I had to resort to code reviews as a means to stop engineers from creating bugs on purpose … I know, sad but true.

But to be clear, this type of “job security” behavior is NOT typical in software engineers process, but it does happen and the PM or team manager needs to get on top of the situation quickly.

Having gone thru Version 4.x and 5.x of Topaz suite of products the bug pattern clearly looks like they are running into the “job security” issue with the development team. I guess they could be very incompetent (I’ve ran into that many times over my 37 years of development) but this pattern is familiar as the bugs are so very obvious they could NOT be missed from even the most basic run the code once type of internal testing.

I hope Topaz can get on top of this, otherwise (and as has been communicated here) people are going to start using other products of similar functionality.