Topaz Video AI Beta v3.3.3.0.b

t9000, it’s not called “play” or “experiment” version: this whole stream is BETA TESTING. I don’t use ANY v3.x in ANY production system. V2.6.4 is the last version that is stable and produces reliable and repeatable results, which is mandatory for a production system.

Others may well have spent more time, but I’ve personally spent over 80 hours fighting bugs, reinstalling, running structured and repeated tests in both v3 and v3b. Both are massively unreliable (especially from week to week). That includes that what you see in Export is almost never as good as what you see in Preview. (It is in v2.6.4)

So, I get a request from Topaz to test Iris and provide feedback, which I was willing to do. But then it asked me to pay to be on testing staff. I ran it in Free mode and it seems to do a potentially good job (especially compared with other models) but with some artefacts worth reporting. But I can’t be sure because there’s a whopping great flickering watermark over the very face I’m trying to see. So I’m not reporting.

Oh, and the one thing I was most hoping for in v3, stabilization, still doesn’t work well despite a truck-load of betas. I recently migrated from Premiere Pro to DaVinci Resolve Studio (plus an RTX4090, yay). Super steep learning curve but so worth it. Its AI stabilization alone immediately produced amazing results with some quite challenging (interlaced) clips right on the first try. So I no longer need or want VEAI v3 for that. And not once since I migrated has Resolve crashed or given me an error of any kind.

So, I’m happy to put in some hours testing and providing beta feedback. But I’m not paying to do so. That’s not “entitlement” as you put it. It’s called “reciprocity”.

Entitlement looks like this: you put out necessarily flaky software and expect those people fighting bugs, furnishing crash log reports, running a range of repeated tests and reporting flaws and artefacts TO PAY YOU to provide that testing service to your company.

That is your choice, but you can’t expect Topaz to change their entire licensing scheme and risking their business for few individuals that are still at v2.x, as most users moved up to v3.x way long time ago leaving v2.x behind.
If Topaz would have delivered Beta versions for free for their Beta testers, they would have to cherry pick very limited amount of people to do beta testing to avoid the situation @t9000 raised. and that would be counterproductive for Topaz as they won’t have sufficient testers available.

As @t9000 said, Beta is an option for licensed risk takers that want to get the latest features before everyone else by “paying” with contribution in the form of tests, feedback, etc…
That is the give & take relationship between Topaz and the End user for the Beta program…

Multiple curious propositions, Akila. 1: Topaz don’t have to “change their entire licensing system”. They USED to extend the license of those who provide useful beta testing feedback. And indeed several of my beta suggestions are now baked into v3. 2. A lot of people (me included) DON’T see beta testing as “risk-taking” as you do. We are more like to see it as exploration and an opportunity to help make things better. 3. Nor are many driven by the ego-need to have “features” “before anyone else”. 4. They’re not “features” until they work properly, which takes quite some time. Until then, they’re ideas or propositions. Some are even dropped or changed before production. 5. You acknowledge beta testers do work for Topaz, in that Topaz would suffer if there weren’t enough of us.

I understand your frustration, and I used to think the same as you did initially long time ago.

But topaz did not invent anything here. it is a common practice in the retail software industry that “beta” program is an option for licensed users that “voluntarily” want to help, it is not a jail-out free pass.
I am on Beta program/tester for other software as well, Microsoft (MSFS), Laminar Research, and others and all are “optional” for those who are licensed that want to take the extra mile “voluntarily” and contribute without any expectation for return or free software, even if it’s in a Beta state (still, it’s a fully functional software after all).

It seems the “live preview” option might be causing some issues. :thinking:
I uploaded an old video and it gave me an error in the live preview section in “export”.
Even when I toggle live preview off, it still gives me an error when I’m previewing it.
I tried to upload the video onto the (not beta) Topaz Video AI 3.3.2 and it works totally fine.
I’m also wondering what this feature is for and why it makes 2X storage space. Is the 2X storage space temporary?

I did a full upscale test export with this one:

  • The weird morphing artifacts don’t still there. I skipped a version as confirmation received it was known and working and I didn’t see a mention if resolved yet, but this confirms not yet.
  • You have changed something on the extract. I still need to put somethign in the name field, instead of it just exporting, but now you start at frame 1 instead of frame 0. Given the input starts at frame 0 which is normal for an image input sequence, this effectively guarantees the output sequence is always off by one frame. Was this somehow an attempt to correct the issues with frame numbers? If it is, maybe step in right direction (?) but unfortunately causes more issues than it might solve, ie:
    – The input target frame is being identified by the input frame number not the export, so do you guess and cut the sequence a frame early or late to try to make the export match?
    – Frame targeting is still off by default. You select 11900 and it starts export at 11899 which means the offset of one counteracts this attempt to rename and you still end up with the wrongly name frames, eg https://i.imgur.com/JjFHJQ4.png this has the input section on left and the output frames on the right. Scene change still on the wrong frame number.

I am not going to do much testing on output quality - its hard to do that when the image moves around like ants, but on a very quick test that I did to confirm that issue, just using Auto, I can see Auto on current Iris is still too soft by default. IE such as this which compares manual settings on last good Iris model to Auto on the current version - Comparisons - Imgsli

Or as an alternative comparison, the same Manual settings in Iris in 3.2.9 and current Beta Comparisons - Imgsli showing that putting aside the werid morphing going on, the newer model is still too soft even manually adjusted +50 sharpen +50 Anti deblur and +60 Details.

I thought maybe this was the new Original Details slider - which was default to 0 - so I tried again at 50% and it made no difference at all, suggesting the slider bar doesn’t really do anything currently. I forgot to name the last image, so its still the frame number, but this comparison is the same as teh last - manual settings identical Iris between 3.2.9 and current, but the named image is 0% Original details the frame number is 50% Original details - More Comparisons - Imgsli if there is genuinely a difference, I cannot see it.

I thought I made a mistake - which is always possible - so I tried again being careful to explicitly separate the images and I THINK I can now see a very small change - Recover Original Details - Imgsli - though it still has lost far too much detail.

5 Likes

The audio options seem to be resulting in enormously large bitrates. Even when selecting specific conversion bitrates. Simply copying also does this.

I’d like to get some info on what error you are seeing as well as the video info & logs.

How To Obtain Your System Profile (Windows)

How To Obtain Your System Profile (Mac)

To gather logs, please select Help > Logging and make sure that there is a check mark next to Logging. Next, recreate the issue and then return to the Help menu and select > Get Logs for Support and attach the zip file to your reply.

the live preview / previews is playing extremely jerky, dont really have time to test more for now, anybody else has this ?

1 Like

Here is the System Report.
Mac mini.spx (6.0 MB)
Here is the logs for support.
logsForSupport.tar.gz (287.7 KB)
Here is a screen shot of the warning in the panel.
Screenshot

1 Like

this is not a bug / issue.
mp4 cannot contain PCM audio track, only mov, mkv or avi can.
your video probably contains PCM audio track and you selected “copy” and export it as mp4 or you decided to transcode (“convert”) to PCM into mp4 file/container

Okay, I just thought it was funny it showed up on the beta version and the not beta version doesn’t complain.

my new pc stopped working after only one day so I am currently a bit handicapped with testing video ai with only an RX Vega 8 on board chip left, but I try to look into the photo tools of course. Switching the license seats worked perfect. :slight_smile:

On my Lenovo Laptop with integrated graphics the mouse movement becomes quite choppy when the integrated graphics unit is working in Topaz. This is a Ryzen 7 5800U CPU with an AMD RX Vega 8 (5000) 2 Ghz. :eyes:

edit : seems that the issue is related to activating the text cursor display and make it bigger in Windows accessibility Parameters.
when turned off, the issue and the jerky playback are gone. Seems it was reproductible in an another software (vegas pro 19) but can’t really be sure, the project I tested is big already and slow down is “espected”. but when i de activated this, the jerky playback in VideoAi was gone immediatly.

don’t know if it’s an issue with Windows 11 (insider preview beta last build) or maybe VideoAi ? anyway, i sent a comment via the windows insider issue report about it.

As @Akila said, you will need to select another container like MKV or use convert