I’ve just tried video stabilization, but it’s excruciatingly slow: 0.5fps on a 4k material with rolling shutter correction enabled (intrestingly, it’s the same speed on Win11 with 3080 Ti and on a 14" MacBook Pro with M1 Pro). Is that expected? I can’t see how can this be usable in the long run 
1 Like
I believe there was some discussion of having facial enhancement back during the beta. As such, I think we should ask Topaz directly. @ida.topazlabs @gregory.maddra
3 Likes
Regardless what it might be named, the result for small faces (an some other small structures) is unacceptable.
As I can see in the videos (TV episodes produced between 1998 until 2021) the problem with small faces is gone for those younger than 2010. That means the older ones must have been encoded in a way which TVAI cannot handle correctly, most obviously in small faces but also in some other small details.
1 Like
The challenge would be to have a face that stays coherent. Try to use Gigapixel AI or Photo AI face refinement on a few frames of a video and you will understand why it’s another story to implement that feature in videos. Also the rendering times would dramatically be reduces by something like 5X-6X judging on the photos software performances with FR on/off.
2 Likes
The same here on Windows 11 , I make shortcut from Topaz Video AI.exe and copy to
start menu by myself.
C\Program Files\Topaz Labs LLC\Topaz Video AI\Topaz Video AI.exe (make a shortcut)
copy to
C\Program Data\Microsoft\Windows\Startmenu\Program\Topaz Video AI(remove original or overwrite)
This is just quick fix, the solution should be provided by developer. As many users feedback, TVAI V3 is far from completion.
1 Like
2020 iMac 27" i9 10 Core, 5700XT 16 GBs, 128 Ram Topaz VE AI 3.0.2, Exporting to G-Raid at 350Mbs write.
Previews not working properly. When I select “Preview” (the Blue button) Topaz doesn’t switch to the Preview window and in the output window below it doesn’t show anything happening unless I switch to all outputs. If I’m on all outputs, nothing is shown to be happening unless I select “Previews”. Such strange behavior for a release such as this.
2 Likes
100% agree
Agree that Chase’s request is a challenge and that it’s hard to do coherent faces but we know that. AI is hard in general but that something is hard and a challenge is irrelevant to the request. Every Pro Editor or Post Production Facility I know (many) would buy Topaz at top dollar if the problem of “monster faces” with bulging eyeballs could be solved. I agree with Chase in that I don’t need a new UI. I need coherence and quality in faces that are not obtainable with Topaz VE AI. Not that you can’t get great results. I use Topaz to great effect depending on the quality of the input. And I realize that some low quality footage may never be workable-garbage in-garbage out-having said that it would be great if on the next series of Betas that Quality is the utmost concern, not new features or UI.
3 Likes
Hello. In v3.0.2: Stabilization is slow with Full Frame option, but fast with Auto-Crop option.
If I try to use the Jitter option, the results are not good with Nvidia GeForce RTX 2060 Super with up-to-date Studio driver. If I use CPU or Intel, the results are good, but the process is much slower. Does not matter if SD or HD video as source.
I like the “drilling down” to the desired model instead of having to pick from a long list.
The text on the interface seems muddy and maybe a too small font size in some places. There’s a symbol to the right of the “next frame” button that doesn’t seem to do anything when clicked. Don’t know what it is for. It’s a little circle with a small symbol inside.
System details:
Windows 11 Home
Dell 8940
Intel(R) Core™ i7-10700K CPU @ 3.80GHz
32GB Memory
10TB Storage / 512GB SSD / 2TB HDD
Nvidia GeForce RTX 2060 Super
1 Like
Still missing avisynth import…
3 Likes
Trim had a preview in 3.0 but if it still works it has none in 3.01? I use trim a lot, but now I need to go back to Handbrake to edit the videos, as the new editing features that were in 3.0 don’t work in 3.01. Sadly for me, I loved 2.64 and the flexibility it and control it had, but the dumbed down UI in 3.0 is much worse and takes more stages to get the same effect. Fortunately for me, I had a backup of 2.64, or I would be left with this new 3.0 which deletes the old 2.64 and I at this point find 3.0 a massive downgrade from 2.64 which made me buy a licence from the trial. In a year or two, I am sure 3.0 will be great, but it’s a buggy mess at the moment. Reminds me of an MMORPG when the devs fix bugs and cause more with the fixes. 2.64 for me and handbrake for the foreseeable future even without updates to 2.64 as 2.64 was and is a great product
1 Like
Face Encoding?
I would think that recognition would depend on skin colors and the perception of facial features. I’m certain that MPEG compression algorithms don’t make exceptions for faces when they are encoding video.
There may have been some change in the encoding standards/methods at that time; they are evolving constantly.
Could the difference be the transition from analog to digital?
The Ui design seems to follow the one from Photo Ai from what i can see. personnally i like it, and i understand that they want a constant between software, as it seems that Topaz want to reduce the availalability of their software to certainly only 2, Video Ai and Photo Ai.
This is a button which is not working properly now because the fix to make the playback of the preview/export going at the normal 1x speed is not yet there. but this small button slow down the playback by 20%. it will be useful certainly when the playback will play normally (not working yet), but for now it’s useless because the playback is already playing slower (3.0) or broken (3.0.1/3.0.2)
I would suggest to most people who have big issues with 3.0.1/3.0.2 to use / try 3.0 instead, lot of bugs which appeared in 3.0.1 and 3.0.2 are not on this version, and the experience is very different.
the 2.6.4 version is available to everybody and can be installed alongside the 3.0
there is an option for the 3.0 to uninstall or NOT the 2.x software during the installation. you just need to not check the button to keep the 2.6.4 version until an acceptable version will be available as suggested in the changelog at the beginning of this thread.
link to 2.6.4 can be find here → 2.6.4 thread.
Situation is a bit difficult for everybody, I hope that a stable version of 3.0 will be there and ok as fast as possible too. Anyway, if people who paid for the upgrade are unhappy, I read they can write to the support to request a refund.
Hello i am new with Video Ai
But i miss a batch processing.
When I add multiple videos, all are processed and then i get an out of memory error.
Working on an MacBookPro M1 Max with 64GB Ram
“… transition from analog to digital …” that can explain the bad quality of input, but that is why I try to recover it by an AI product.
I agree that output quality should be the first priority. The fact that the “Quality mismatch between 2.6.4. and 3.0.x for some models” (which has existed for a long time in the betas) is still an issue in this version is not a good sign regarding Topaz’s priorities, in my opinion. I realize that Topaz needed to develop a new Video AI, and I think Photo AI was inevitable (I even speculated on such a combination app before PAI’s official announcement), but for much of the past year it really seems like the upscaling AI models have been placed on the back burner.
The face recognition in Gigapixel and Photo AI is excellent for small faces but does not work well for large faces (and in fact the official recommendation is to turn it off for large faces). Since the same face can be of different sizes in different parts of a video, that would mean Topaz needs one of two modifications:
- Modify the face recognition feature to work on faces of arbitrary size. This option is probably not a good idea since the usual upscaling models work well for large faces.
- Activate the face recognition feature for faces that are smaller than a certain threshold. To avoid sharp transitions as the size of a face changes, for faces slightly larger than the threshold it could upscale the face with and without the face recognition, and then blend the two.
1 Like
For me – VEAI 2.6.4 seems to be the best option available. I can stabilize video in FCPX or AE (so I don’t care about the feature). I can de-interlace with Compressor (so I don’t need this software to do it). This software filled a void of video enhancement that you could not find anywhere else.
No sense in re-inventing the wheel on some features. I hope the software stays more focused, with fewer features – but features that do an excellent job.
For what I do it takes .09 seconds per frame and the results are very nice – using Mac Mini M1 16GB.
I hope Topaz will continue to support v2.6.4 (not expecting any changes) – just for it to continue to function. That is my worry.
When Version 3 is better in performance than 2.6.4 – then I will be glad to upgrade.
Regarding face recognition in video – that is going to be tough to offer at a speed that is acceptable. I don’t see that option being available in the near future. I’d rather just have fuzzy video on faces.
Now - these forums are awful to navigate. They aren’t support forums – they are bug forums.
I hope at least these can be cleaned up. Beta testers should have their own private forum.
4 Likes
There is no beta version of Version 3 now that it is a released product. That is why you see beta testers replying to this forum. They are labelled as beta testers here because they are registered as beta testers of any beta version. When version 3 was still in beta stage the beta testers did have their own forum for each beta release.
1 Like
We’re aware of the performance regression and it’s one of our top priorities to address. As for the license expiration, if your license covers v3.0.0, you’ll be eligible for any v3.0.x version.
7 Likes
