Topaz Video AI v3.0.10

This was one of the big complaints with Version 1 and 2 was when they crashed you lost everything and there was no hope of recovery. The GUI(including preview) and processing engine had to be separated to get around this. That was why many got upset when Trim lost frame numbers. It was easy to divide a video into sections by frame count and then recover only the section(s) you lost after a crash.


In VEAI (v1 & v2) the GUI .exe was the engine too; come TVAI (v3), they apparently moved the engine entirely to ffmpeg. Check! Since the latter is now the engine, though, makes one wonder what the difference is then, crashing-wise (presuming it was always the engine crashing). Was it the GUI? Was it Preview causing a crash? (If so, they now elegantly solved that by giving us only a half-working Preview. :joy: What, too soon?).

I now see where you’re coming from, though, with Trim not having accurate frame numbers.

1 Like

You really have to do more extensive facial expressions and clothing model’s training.

1 Like

Guys I am new to Topaz Video AI, whenever I try to do some changes, it gives me this error message - any idea? I am on a MacBook Air M2.

Thanks in advance!

1 Like

Thanks, I’ll try DDV2. Did you test DTV for ghosting, as well?

The current timeline trim is hit and miss for me. I trimmed a video, and the export was entirely a different trim. On the second try, it seemed to work fine.

An update: I did a real test on a real video I had recorded on a Sony A7S II, initially processed with Avidemux 2.8 for some trimming and color & brightness adjustment and saved to an X264 encoded mp4 file – after chugging along for couple of days, the end result was nothing but a still frame of some part of the video. I think I’ll have to uninstall and wipe up all traces of both the Intel drivers and Video AI and do a complete reinstall using the last previous versions that worked.

Yes, I forgot to add that only DDv2 worked. What does these lines mean?
PS Have analysed in detail output quality?

1 Like

Thanks for the contribution.
It would be interesting to compare with the two-process method.
01 that you already made + 02 Interlace Progressive Dione Robust 1x.
With the image you propose is a conclusive test.

1 Like

Except for the few times when an FFmpeg ‘temp’ file has been completed; that is with valid ‘header’ information, I assumed that all of the MediaInfo was furnished when the file was passed back to the GUI from FFmpeg and the lid renamed to the ‘targeted’ name.

At this point, by reading the FFmpeg command string the GUI passes to It is obvious that the GUI isn’t given the final output name, so it could be possible that not all of the GUI’s functionality disappears w8th a GUI crash.

Finally, I think the complexity of passing jobs FFmpeg jobs form and then back to the GUI, may still be problematic, which may explain the limit to the number of simultaneous jobs the GUI can invoke. - Presently, selecting all source files and then pasting the same configuration to them will result in an "Export #'. The # represents the number of selections. It gives no destination file name, but starts exporting immediately. When attempting to export this way, several of the spawned FFmpeg jobs will fail, but would otherwise complete correctly if run singularly.

So, until this multiple file export scenario becomes more stable, I’m going to continue Exporting one or two files at a time, starting each one individually.


That all depends. There are a lot of variations in what is commonly call ‘interlaced.’ And other forum contributors have mentioned there are both framing and temporal considerations involved.

Another bugaboo is that it appears that some optical media video was purposely re-interlaced so it would be playable on both newer or older systems. -

So, the answer to your question is that presently, there is no be-all and end-all solution for deinterlacing.

Addendum: A great deal of the success or failure in doing good deinterlacing comes from being able to identify the ‘archeology’ of when and how the source material originated, whether it was originally created on film or videotape, whether it was interlaced for broadcast/viewing before or after the popular transition from SD to HD and various related factors. - The last and one of the most important other considerations is to be able to deinterlace without creating additional noise and other artifacts.


I agree. With interlaced video most if the problem is truly identifying the type interlacing. All one has to do is look at the number of different methods for deinterlacing to know there is no one way. Looking at Staxrip as an example and the number of AVS filters for field manipulation. That’s enough to make some get dizzy especially since many can have the parameter values changed.


Summarize what you have found

・When processing interlaced video, results are the same for MPEG2-TS-1440x1080i, MPEG2-1440x1080i and MP4H264-1440x1080i.
・Because DDV has better image quality than Proteus, if you deinterlace it with other software and process it with TVAI, the image quality will drop.
・Medium image quality such as DVD-720x480i video requires further verification, but there is no problem with DDV3-3.0.10.
・VEAI2.4.0 VEAI2.6.4 TVAI3.0.0-5 TVAI3.0.5 works with a single folder, but TVAI3.0.10 should be installed normally.
・TVAI3.0.0 to TVAI3.0.4 have an adverse effect on the image quality of VEAI2.6.4, so be careful. Installing TVAI3.0.5 or later will improve it.
・If you use Chronos-300% on the processed video, you can visually confirm the defects in the video.

--------------Output MPEG2-TS-1440x1080i with each version--------------

DDV2-2.4.0 3840x2160 normal output
・You can’t use long videos because they sometimes drop at the end of the video.
・You can reliably output with Cancel → Yes just before the process ends

・DDV3 cannot be used
・Deinterlacing is the same as DDV3-2.6.4, but there is no afterimage.
・Chronos-300% is fine for anything other than fast-moving videos

DDV3-2.6.4 3840x2160 normal output
・The image quality is the best, but there are deinterlaced parts and afterimages
・When using Chronos-300%, afterimages appear on all movements

DDV3-3.0.5 3840x2160 normal output
・Deinterlacing is good, but there is noise overall

DDV3-3.0.10 3840x2160 normal output
・Same as DDV3-3.0.5

DDV-3.0.0-5 2880x2160 (Early Access) normal output
・It can be output in the size of 1440x1080 of the original video.
・Image quality is not as good as DDV3-2.6.4, but good deinterlacing noise removal
・Chronos-300% stable for all videos

Videos created with TVAI3.0.0-5 give the best results when using Chronos-300% with these versions


I wrote a summary, so please take a look.

I wrote a summary, so please take a look.

I wrote a summary, so please take a look.

I’ve been running 3.x on my Macbook and criticizing the performance regression several times, but nowadays I find it just fine. Don’t know which version is the turning point, but performance wise I finally don’t mind using 3.x over 2.6.4 at all! Great works, guys!

Nice! Does Staxrip output 2x fps on all those alternatives?

No clue since I don’t do 2x frame rates. I actually reduce high frame rate videos. I would think it would as AVS has many filters. I do know from experimenting with QTGMC if you remove the parameters and just use QTGMC() you well get double the frame rate. I would think Hybrid and some of the other full featured encoder/converter software that work with AVS can do the same things as Staxrip if you dig into them.


A newer version is available.