Video AI 3.0.0.0.b

This is PhotoAI.

https://community.topazlabs.com/c/releases/topaz-photo-ai/85

1 Like

Thanks , was not aware of this !

1 Like

Yes, that’s on a MacBook Air M1 (Monterey 12.5).

And yes, that build works just fine
being labeled Alpha means that it is a previous build from the Beta? What am I not getting with the Alpha build that is now in the Beta build?

I am not at home until Monday. I already uploaded it to the Dropbox folder which is linked at the top of this thread.

1 Like

maybe they’ll put the software back in alpha ? no idea, looking forward to the answer too ;-).

Some folks at the Facebook Topaz VEAI group have managed to use the Avisynth filter chain with v2.xx. They did that with exactly that purpose to feed deinterlaced, properly cropped, and so on video without reencoding degradation. VEAI v3 will use internal FFMpeg filters for that purpose as I understand it.

1 Like

In v3.0.0.0, the core of VEAI is now ffmpeg and can be processed on the command line, will this continue in the official version?

The advantage of ffmpeg is that it can be processed with filter chains and batch files.

Setting up a filter chain on the command line gives you more flexibility in combining filters than GUI presets.

Also, when processing multiple files, if you create a batch file, you can simply drag and drop files into the batch file and the process will proceed at once.

Yes !

1 Like

Is there a way from the GUI for outputs to have CPU h264 encoding? The only h264 encoding selection is through the GPU, NVidia or AMD depending on the card in the machine.

beta Is not Producing H265 Main10 According to MediaInfo Program.

Do you know whether your machine has any hardware capable of h.264? If it doesn’t, then the file would need to be encoded via software, which could be very slow. FYI: some encoding modes may or not be available, depending on the bit depth you’re looking for.

I had a problem using the software to double the frame rate of a 4k video that had already been encoded with these settings and found that using that setting again created a video that was skipping multiple frames within.

If your replying to me I should have said CPU software encoding(h264 via FFMPEG) instead of just CPU. I use a Ryzen 9 3900x and have no problem with software based encoding speeds. Even 4K isn’t that bad for my needs with the 3900x. I just like the quality of a software encode for some videos. Anyway it would be nice to have that option back like it was in older versions of VEAI.

If you were not replying to me sorry.

1 Like

“No accelerated colorspace conversion found from yuv420p to bgr48le.”

I fixed this by removing the custom paths, and went to defaults. My c: drive doesn’t like it, but it works.

Fix:

Something rather odd occurred today. Yesterday, I rendered an interlaced input file to a progressive mpg using v3.0b, Dione Robust, Artemis denoise and sharpen, no resize. It did a beautiful job but ran painfully slow.

I attempted to repeat the process today on a nearly identical file and the preview started and hung. I closed everything and tried to duplicate the job I did successfully the previous day using the original file. Same result, nothing happened.

I noticed on the Nvidia website that they had a 3090 driver that was a month newer than the one I had. I downloaded and updated the video driver, It made no difference whatever in v3.0b’s behavior.

I fully uninstalled VEAI 3.0b deleted the temp information and removed the registry entry “Computer\HKEY_CURRENT_USER\Software\Topaz Labs LLC\Topaz Video AI BETA” . After reinstalling v3.0b I opened preferences and preferences > AI Processor field was blank and completely empty, and so was the Output> Encoder field.

I repeated the process and still nothing. I’m stumped. I don’t know why the beta stopped working between yesterday and today. And I think my attempt to reinstall simply exacerbated the problem.

Is it possible there was a ‘leftover’ artifact from v3,0,0,40a that got removed and was omitted in the install for 3.0.0.0.b?

Any suggestions?

2 Likes

dont’ take what i tell for 100% solution, just saying it like that.
i’m afraid that at this point, you’re good to do a full complete clean uninstall of the 3.0 version. something went wrong somewhere, and until a dev or support help here on the beta, is able to reproduce it or see what’s wrong in the log, don’t see other possibility.

for me the 2.6.4 chronos is not working anymore since two weeks (everything else does).
fortunatly this feature work in the 3.0b lol.

There is an error in the log but i’m too lazy to write to the support or doing the clean install (at we must erase everything, from the registrery to user/windows folder etc. i’ll see later, but certainly, a full clean / reinstall is certainly the solution, until it’s related to the graphic driver. on the website 3Dguru, they have if i remember a full_clean_remove procedure for nvidia driver when one is corrupt, but not sure it’s still available. i think Nvidia lately maybe did their own clean software and so their solution is useless since then, but i’m telling you all this without being sure at all.

A couple of days ago there was question of testing multiple dissimilar GPUs. Here is a quick test I ran this evening using a RTX 3060 12GB, RX 6600 8GB and the IGPU in the 8 core/16 thread AMD 5700G CPU. The system has 32Gb of ram.

This is a simple comparison of multiple Graphics Cards and IGPU usage in both VEAI 2.6.4 and VEAI Beta 3.0.0.0.b. This is both the single GPU and then using the ‘All’ selection. The test was performed with a video of 854x480 converting to 1920x1080. I used ProteusV3 estimating values from the first frame. The numbers are rounded.

VEAI 2.6.4
5700G .41sec/f-2.44fps
3060 .13sec/f-7.69fps
6600 .17sec/f-5.88fps
All .22sec/f-4.55fps

Beta 3.0.0.0.b
5700G .53sec/f-1.9fps
3060 .24sec/f-4.1fps
6600 .37sec/f-2.7fps
All .17sec/f-5.7fps

Here are my conclusions. The GPU’s are being much better utilized in 2.6.4. Simple wattage comparisons from OpenHardwareMonitor show for the 3060 in 2.6.4 power usage of 105W compared to 48w in Beta 3.0b. For the 6600 2.6.4 was 69w compared to 25W in Beta3.0b.

In the ‘All’ case it is clear why Topaz labeled ‘All’ as Experimental in 2.6.4. It does not work properly in mixed brands. In Beta 3.0b it does work but the processing from a single GPU in 2.6.4 was still faster. When Topaz gets the GPU optimizations done in 3.0 than multiple GPUs will be a very good feature even if you only have some slightly older GPU’s laying around and as long as you have a power supply that can handle the total power.

I still personally think the ‘All’ selection should be replaced with check boxes to allow selection of GPU’s. It is possible to have 3 or more graphics processors in a single computer. Threadripper/Epic systems come to mind along with what might become very low cost soon, ex-mining rigs. You may not want ‘All’ selected.

One last note about the hardware during the test. This experiment was done on a B550 motherboard with mixed PCI slots, PCIe 4.0 x16 and PCIe 3.0 x16 (x4 mode). The 3060 was in the 4.0 slot. The 6600 in the 3.0 slot. The back of the 6600 was maybe 6-8mm from the fans on the 3060. I never had anything but normal temps while testing but the case side panel was open. Ambient temperature though was high(for most people) at 27C. I’m in Florida.

4 Likes

There is no info what the software does. No processing output preview, and also it sits for 0% for a long time (only ctrl+shift+esc, and checking out network usage you can tell it’s downloading a model). 3.0 doesn’t give you a hint of what it does.

If I choose only double the frame rate, what conversion does it use?

If Proteus is selected. Why is everything set to zero, why not remove the sliders completely when the setting is auto?

Where is the grain?

Audio has an option for copy and transcode. If I select transcode, what does it do? There are no options for that.

1 Like

you must activate the model, so appolo, chronos fast or normal. yes, it would be better to have these two filters reunited in one.

from what i understood, as it has been explained before, there is 3 possibility to use proteus

  • the auto feature calculate the setting for each picture. the slider are at zero to allow the possibility to add or remove a % of setting on a specific feature. so let’s tell that you think that in auto, the picture is not sharp enough, you can boost it by +30, or in the reverse, it’s “too sharp” you can remove a"-50%"

  • you have the estimate fonction, which calculate the setting from the picture actually selected (on the timeline). it’s the same feature than in VEAI 2.6.x

  • if you turn off the “auto” mode, and don’t use the estimate one, you have the “normal” proteus feature, aka, the settings you use and put yourself.

if you ask yourself why it’s not explained anywhere, it’s because they didn’t added yet to the software all the tooltips etc, to the GUI.

feature certainly added later. as simple as that


in a perfect world (not today), copy would be a passtrough copy of the audio track present in the original file. so if you open a mkv with a 5.1 ac3 640 kbs file, the final ouptut must have the same audiotrack with your video processed by VEAI.

the encode option is certainly a possibility to re encode the audio track
 so there will be certainly an option somewhere or at a moment, where you’ll be able to set up the bitrate of the audio track and the format.

it’s evident that if these 2 features are not working properly yet, it’s because it’s buggy or still in developpement and not yet available for final use.

If I don’t know what the functions do. How can one do any testing?

I’ve seen this movie before. Once there was a very similar software to VEAI called vReveal. It used CUDA. It was a real game changer to cellphone videos etc. When they went to Version 3, the software died. With vReveal 3 they removed functions like grain, and made completely useless changes. Too bad you can’t use the vReveal anymore, because you needed an online connection for that. I would be still using the software today. It was the only software that fixed micro jitter perfectly. Zoom ratio for stabilization. Plenty of great features. With version 3 they made the software simple and removed options that a user could modify.

If this would be my software. I would add the stabilization and Apollo models to Version 2.xx. Then continue with Version 3. This is so far from the finished product.

For those who wonder what vReveal was (when it was still good).
https://www.overclockersclub.com/reviews/v_reveal/

2 Likes