I have been using Topaz Video AI and Topaz Video for a almost a year now, and after the big update where they announced their new Topaz Video and Subscription Options a couple months ago Starlight Mini Renders have slowed to an absolute crawl for me. I had it down to exports for 4K to be around a 1 hour for every 20 seconds or so of video which I already thought was pretty long. But, now the past couple months I can’t use Starlight Mini because it is taking 2-3 days for around 45 secs of video. I chop them up into 45 sec to 1 min long videos and let them render overnight and while I am at work but still they take days. I am ensuring I utilize my GPU and not the CPU and allowing max allocation to the app while rendering. Anyone else having this issue?
I can’t begin to speculate why this has happened to you, and I also can’t say that I’ve seen a similar slowdown, but…
Users who’ve not read this article on the support pages may not be aware of the shell game that Topaz Labs has played since the original Project Starlight appeared earlier in the year as an online-only service. There have been confusing changes that impact not only its use in the cloud and/or locally, but the quality of the rendered video files. These changes weren’t clearly articulated so that we could understand and follow along.
My first experiences with Project Starlight were the weekly 10-second previews that we were allotted when it first came out. I was blown away at the results at times, although there were also some troublesome areas, like mangled text and facial distortion on smaller subjects. But I was thrilled at the results and looked forward to the day that I could run this locally on my - then - AMD 7900XT with 20GB VRAM GPU, which they promised to eventually support. (We now know how THAT worked out. But I digress…) These early experiences generally took from 30 - 45 minutes to finish up on the cloud servers, but some of the things that were cleaned up were remarkable, IMO.
Then, TL announced a speedup by a factor of 10 on the cloud-based Project Starlight. Sure enough, it was now taking 3 to 5 minutes for my 10-second preview clips to finish. Cool!
Not long after that, we started hearing the name ‘Starlight Mini’, at about the same time that TL announce it would be available to run locally on nVidia GPUs. At that time, I choked and bought a 5070Ti because it wouldn’t run on my AMD GPU. But I could at least now run it on my system. The naming became confusing to me, because it seemed that Starlight Mini was to be the name of the model running locally, but they were also using it to refer to the cloud version, which had been referred to as Project Starlight.
So Starlight Mini started showing up in Video AI 7, and in new beta versions. And support for AMD & MAC GPUs also appeared, with immediate howls from those who had been using Starlight Mini on nVidia GPUs, protesting that the quality of the new model was inferior to what it had been in Video AI 7. TL acknowledged this as being accurate, but temporary, as this was just an initial release in the beta to see how the AMD and MAC users’ fared. It would improve, they said. I wondered why the nVidia model had been degraded - aren’t there different models for AMD & MAC - they can’t run the exact same code, can they? At the same time, a new Starlight model - Starlight Sharp - became available for testing. Cool! And it was a bit faster than Starlight Mini. Even better, at least in terms of waiting time.
Then came the subscription transition fiasco and now we find out that AMD & MAC GPU support is a ‘Pro’ feature. (Thank goodness for ‘founding member’ status…) But the new Topaz Video (Studio) app no longer had a enhancement called Starlight Mini. It offered “Starlight” and Starlight Sharp. What happened to the ‘mini’ part of the name? And what of the online model that started out in life as Project Starlight, but later seemed to be referred to after some time as Starlight Mini in community posts, but wasn’t called that when you sent a file to the cloud to render?
Somewhere along the line, a new name appeared: Astra. This was something new that we didn’t have access to until our accounts transitioned, but it required credits to run.
If you read the article that I linked to early in my post, you’ll learn what happened to the original Project Starlight that produced such wonderful results in my free previews. To quote from the article: “This specific AI Model has since been renamed as Precise.” In other words, the model that made me enthusiastic to be able to run ‘Starlight Mini’ on my local system was move to a pay-tier and replaced with something with a dumbed-down model capable of running on my computer, not one that has been ‘optimized’.
With this name shuffling going on - with what we thought would be Starlight Mini in our new Topaz Video app now called, simply, Starlight, and what I thought would be Starlight in the cloud version now moved to Astra and called ‘Precise’, what do you get when you select Starlight as the enhancement and export it to the cloud? Is it the same ‘Starlight’ model that runs locally, just produces a file more quickly? Or is it any better in quality?
I don’t know. But what I do know is that Topaz Labs’ strength is in producing software and confusion, not in clearly communicating what is going on.
The devs are looking into issues with processing rates and working on making them more efficient. There are several backend things that are hampering this but they are working on them as quickly as they can.
Your notes on the situation and changes have been shared with the product team so that they can work on being a bit less confusing in the future about changes to the app. Several aspects have had to shift as the devs have been working to improve and update the app and some of these have not come out as cleanly as they would like.
Thank you for passing this along. I can’t speak for other members of the community, but I’ve experienced a lot of confusion and frustration since mid-August, trying to keep up with changes to Video AI 7.n, the Video AI 7 betas and the transition to Topaz Video.
Although it’s not pertinent to this topic, I’m encouraged that my first significant test of Topaz Video 1.03 shows that there has been a welcome fix regarding output resolutions and sneaking a look at the export preview window is looking pretty good, too! I’ll know more tomorrow when the export finishes.
