When I went from 1.2.1 to 1.3.0, Starlight Mini got slower (0.3 fps rather than 0.4 for my current project). It also produced significantly larger output files and generated more obvious tiling artefacts. I switched back pretty quickly.
Let’s continue Starlight HQ tests.
This time the input is just horrible. Old footage badly upscaled to 1920x1080.
This is what I’m working with.
First things first, downscale to 422x308 and upscale with SeedVR2:
Not much, but a ground for Starlight.
Starlight HQ:
Not so much gained.
Starlight Mini:
Oh, a face!
Now let’s add Iris on top and some grain to mask flat surfaces.
Here are the videos:
After SeedVR:
https://youtu.be/dNEYxd0NpPQ
Starlight HQ:
Starlight Mini:
Starlight Mini+Iris and grain:
In the Part 2 I’ll share my experience with the architecture shots.
Starlight HQ and architecture.
The input is good, 720x576 50i, deinterlaced in Hybrid to 768x576 50p.
2x upscale, Proteus Natural for a reference:
Not so bad, considering fast speed.
Starlight HQ:
Ugh. Look at the diagonal lines and the windows. It’s like oversharpened old Proteus with artefacts.
Starlight Mini:
Now that’s really good. We can see even a texture on the windows drapes.
I’m out with Starlight HQ. I don’t see where it could be useful.
Here is the original:
Starlight HQ:
Starlight Mini:
I see that local Starlight Fast 2 and Starlight HQ are for Nvidia only. Is Mac support being worked on? Or will this forever be Nvidia only?
I’m genuinely perplexed by Starlight HQ. A diffusion model that only produces decent results on HD-or-better material? Baffling. Starlight Fast is a much more sensible development of the Starlight model but of course I can’t use it (not enough VRAM).
The Starlight HQ model can be used on other files with good results as well. Case in point screenshots from an older file in 768x576 resolution. While it was trained and designed for certain types of videos and not as ideal for lower resolution file, it can still produce very usable results when doing upscales to 1080.
Also to clarify this was on upscaled to the minimal factor.
The visible blockiness in the foreground of the ship output is pretty much what I was getting in my tests with sub-HD material. The model seems to struggle with low-res scenes that are dark and compressed (always the hardest scenes to fix, of course). I’d be interested to know what kind of material it was trained on if you don’t mind sharing.
I have written to TOPAZ Support for LOCAL Access to Starlight HQ and Starlight Fast 2 Local, and previously several FOUNDERS have written here on this same matter, yet STILL nothing about access has progressed after several days - despite “Look forward to an update in the next couple of days!”. Unacceptable originally, now insulting to FOUNDERS.
Agreed
I also would like to add my voice to the rest of the founders here. Although I appreciate the hard work on Topaz video but the founders should always be a priority hence why they are founders. Not being able to access to models upon release defeats the purpose of being a founder account owner. Happy Easter everyone.
agreed
Hello all, just stepping in for an update here. The team has spent the entire week working on the Founder issue and several other bugs that were somewhat adjucent in the code to try and get this patch together and released. They are still working on it now but getting very close and the support team really appreciates everyone for their patience. We understand that desire to test out the new models and the frustration over the terminology and wording that led to a lot of confusion with the access internally with the company as well as for users.
As soon as they have this patch build put together and ready for rollout they will be pushing it out.
The attitudes you are seeing in this forum shows the value of work logs. When you have what amounts to a major issue for a substantial number of users, the worst thing you can say about it is nothing. You could have headed off a lot of the hostility you’re getting with a list of the bugs being worked on that gets items checked off daily as they’re resolved.
Within the last few working days, the team has addressed the Founders issues in addition to some other bugs to give everyone a better experience. We will be doing internal QC to ensure these things will be working once updated. We estimate that this will be available early to mid next week once these internal checks pass. We appreciate your patience and understanding for this quick turnaround. Thank you
Don’t give up without a fight, once you lose that FOUNDERS status (and TOPAZ are being held to their word, believe me) you will be paying for PRO or on CLOUD for the new models.
Thank you for this long awaited update.
Yeah, luckily, I bought the 2nd product right before the switch, so I am good until then. I plan to keep mine active again, come that time.
Thank you, while this has been frustrating, it is nice to see the company making this right - and appreciate the relatively quick speed this is happening at. (Understanably some of us are impatient and want it yesterday, but being familiar with development, we also need to see the other side of this)
HI, Is it just me, or is Topaz getting slower and slower? I still can’t use any Starlight templates, and now it’s getting slow. It’s really frustrating. Does anyone else have this problem? I’m using a Mac M4 Pro.
On Windows not, I get some increased speed in V1.3. Model SLM 3x from 640x480 to 1920x1440 → 0.7fps (RTX509 @850mV undervolted), SLS and other models not tested yet. I’m happy, but they really should start paying more attention to Mac users.











