link is wrong, it goes to v6.1.0.4b instead of 6.1.4.0b
it’s also in description wrong.
when launched it shows that there are updates and 5000 opt dont work, so def wrong installer.
Update 1:
now filename is correct but directory in link is still wrong: 6.1.0.4.b/TopazVideoAIBeta-6.1.4.0.b.msi
Update 2:
downloading with manually fixed link
Update 3:
that version says it’s 6.1.4 but it doesn’t contain RTX 5000 opt
I let installer clean old model versions. Testing was done with benchmark, which gives low scores pre 5000 opt.
Ok, that’s too much for tonight (Europe). See u tomorrow with hopefully a correct and working version. Night!
Tony will Topaz get better speed optimization for M3 Ultra? Has the team thought about getting the 512GB Ram version of M3 Ultra to see if Starlight can utilize that massive amount of RAM for the large language AI model?
In this beta I’ve noticed the program doesn’t download and use tensor models for RTX 3090 anymore, are they being prepared in accordance to the 5000 series again?
Those are the models that downloaded themselves while me just previewing with GPU set on 3090 first, then on Auto.
Ah, this beta had TEMP folder wipe bug and preview files being created next to the original file, as usual.
Roll back to a stable verison for the 30 series if it’s causing issue in the beta.
50 series optimizations. man… I was going to buy a 5090 at launch and decided to stick with my 4090s because they just seem more reliable right now, but the 5090does seem like it COULD way more than 30% outperform a 4090 at tasks like topaz with gpu heavy modeling.
Any idea if the motion flickering/ghosting in fast movement in Starlight is being caused by the compression of H.264? Hope we can get some more export options when using Starlight soon so I can see if that’s the problem. So far I’m not seeing any of these flickering, ghosting/blurring motion wierdness in any of the older models.
I mentioned that this beta totally ignores tensor-optimized model files for 3090 and downloads .ox ones only, but there’s more - the program errors, when I put the .rt806-8500 models into the models directory and try running a preview:
The sample renders we’re allowed to do are just not big enough to even consider justifying the cost of paying for credits on a huge gamble using starlight IMO …
I realize there are price issues going on here, but the fact is that it costs a LOT of money just to do 5 minutes worth of stuff with starlight and a 10-second (max… framerates often lower this) sample is never enough to judge what it’s going to do to an entire video.
I would consider using starlight if i had more confidence in it at all. It 100% has fixed some issues with videos where I’ve had no success with other methods (topaz models + a litany of other software suites with ultra-robust plugins… I’d just name the synth plugins plus borisFX stuff to start… not to mention everything i have for Vegas and other programs)
Starlight could be viable for some projects I’m working on , but I’ll never know, because the price to gamble with “maybe” is still way too high.
It seems like by the time they make starlight viable at al for people that other companies (and freeware tools) will be quickly catching up. There’s already plenty of ways i’ve implemented stable diffusion with a 4090-upscaling using unorthodox workflows that pretty much do the same thing.
I really thought topaz on end user machines was making a lot of progress with rhea on things that topaz couldn’t’ do with previous models, but it seems they just kind of abandoned it to focus on a product that isn’t’ getting adequate testing (unless there’s a select group of users being given a lot more acccess to test the modeling capabilities)
I’ve used topaz on tens of thousands of hours of footage at this point for various purposes… mostly upscaling and using models more as tools than as a "magic wand with pretty great results, but iit just seems like the cost of the programn has ballooned year-over-year (hell i used to get free or discounted subscriptions just because i did so much testing, but those employees left and the new guard doesn’t seem to care at all how much money and time people put into testing their stuff… there’s no reward system anymore like there used to be for sacrificing your personal time, computing power, and money it takes to do it all.
I realize these guys are busy, but they used to at least keep in touch with their beta testing community and respond to the helpful feedback fairly promptly… Now they just post stuff and go “eh whatever, release it with the bugs and make the beta testers pay the new inflated prices for a program that isn’t being improved for local processing at all”
Really hope we can get an update on these motion artifacts and if they’re tweaking the Starlight model to address some of the IQ issues I’m having. I agree with you, and I think the Topaz team is just very busy which is maybe why they’re a little more quiet. I’m sure they will respond to you if you shoot them an email.
