Thanks for sharing your similar experience.
I was able to get my M1 Max to preview without black screen, but it errors out when I try exporting my 20 second clip.
So we have two reports of M3 not working, one of a M1 based rig which did partly work and my M2 that works.
M4 is missing but I bet itâll work as well.
And, probably related, M3 based Macs were those with the most issues with TL software in the past, not only TVAI but also TPAI and TPGP.
In any case Mac users where SL mini doesnât work donât lose much ATM as the (actual) quality really isnât worth the wait.
1 Like
It says âAdditional Resolution Optionsâ for Starlight Sharp, but I only see Original Size, 1080p, and 4K. Is htat intentional? If so, that seems like a pretty massive step backwards. Now Itâs going to bake in black bars for anything thatâs not 16:9, and I donât have the option to choose a specific rendering resolution in-between âOriginalâ and 1080p in cases where I might want to save some time.
Beyond the quality comments for SL Mini, I"m also confused by the change in resolution scaling options there as well. As mentinoed above, having the 3x upscale option was quite helpful.
I was also getting wildly incorrect fps readings for both Nyx XL and Starlight Mini.
On an RTX Pro 6000 Blackwell and 9950X3D: Nyx XL at 4K - it started supposedly at 2 fps, then increased slowly up to 8 fps according to the estimate. It actually took 5 minutes to denoise 300 frames (after model was loaded), so basically exactly 1 fps. I thought something might have been wrong, so I completely uninstalled, deleted the ProgramData folder and reinstalled, and nothing changed.
From my limited testing (havenât run that many frames through it since Iâm trying several things and given how slow it is), it seems much more likely to break and be overly smooth / blurry compared to a lot of the other models.
I had this happen on a DVD quality video with quite a lot of noise, so I tried a UHD Bluray quality example as well, shown below:
Original Frame (re-encoded with exact same settings as the other examples below):
Nyx XL:
Nyx3:
In just this one short scene, Nyx XL blurs and unblurs on her forehead 4 separate times. Regular Nyx is certainly less aggressive, but it doesnât break during this scene.
3 Likes
Question: multi GPU AI processing is currently a Pro-only feature. Will the Studio apps support that, will it still be behind an additional paywall, or will it go away altogether?
Would like to know this before I decide on a GPU upgrade.
2 Likes
I have to correct myself about speed (wrote that the new Starlight-Mini is much faster):
The new starlight shows the wrong Speed while rendering.
Upscaling 480x360 => 1280x960 old Starlight Mini: about 0.7 fps
New Starlight Mini: about 0.8 fps.
(But the quality of the new is horror!)
The thing what is giving more speed is the option to upscale to lower resolutions, for example 320x240 > 640x480.
5 Likes
I donât know what the difference is or was, from the past tries with Slm on my part, but I finally got it working minutes ago. I noticed that there is a Pause option now, which is great!
Whatâs up with the Focus fix in there also? I tried that and all I got was a video that was placed up in the upper left hand corner of the player.
Would it be possible to have SLM just leave text or faces blurry if it canât make them out? It can definitely tell that there is text or a face there since it tries to fill in the blanks. I rather have blurry text than alien characters there instead.
1 Like
I had it do that once too. I think I set it to FHD output instead of X2 or X3.
1 Like
I donât understand the logic here is putting out a dummy down version of starlite mini to test on the macs. I like to beta test with actual content that I am working with, because as long as it takes to encode with SL mini, I donât have time to beta test without making good use of that time by actually testing on working projects. That being said, just give us the full potential of starlite mini available on the mac as it was before this test and I will be happy to test it out. But it seems kind of pointless to waste time testing on material that will be pretty much useless since the quality is so bad compared to what it was doing previous.
2 Likes
The idea is to first make sure it doesnât crash a lot. Optimizing software that canât even finish a test run is an exercise in futility.
1 Like
Hallöchen 
OMG, SLM output looks practically the same as the input. Applies to nvidia users too atm, I guess they will fix Starlight in the next beta version which will be released.
Have you tried using Nyx XL as a second enhancement only (in conjunction with Proteus for example) instead of using it solo as main enhancement ?
In protest against recent changes:
- Starlight releasing as a cloud only model in a app that was focused on local processing for many years.
- Starlight Mini releasing as a NVIDIA only model, and now AMD and Apple support being locked behind the Pro license.
- There is also no mention of Intel GPU support. Although I havenât tested my Intel GPU in Topaz Labs Video Enhance AI in a while, so I donât know if Intel GPUs are even still supported.
- Starlight Mini shifting from the âstandardâ license to the Pro license.
- Starlight Sharp being a Pro license model.
- The Pro license being a downgrade over the âstandardâ license in some aspects while costing significantly more.
- Along with many other not great things about Topaz Labs Video Enhance AI that Iâve seen over the last few years.
I will not be providing feedback on this Beta, or likely future Betas.
9 Likes
You forgot the quality downgrade of SL mini there.
I somehow fear that SL mini was too good and now they have to cripple it down to be able to sell SL âsharpâ.
3 Likes
Yes, atm SL mini really isnât worth it.
And, Iâm not quite sure why with the first iteration of the model on another platform the quality should suck so bad and even more so why the quality on the already existing platform degraded that much?
Itâs not as if the model itself wasnât there for quite some time. Slow speed, stability issues and high RAM consumption I could understand but not this âqualityâ (I hardly can type this term seeing the current results of SLM).
Also what are those ongoing problems with M3 chips (TVAI but also e.g. Redefine) - since months they donât seem to be able - or willing - to solve that.
1 Like
I havenât, did you have a specific use case/ workflow in mind?
For me, at least when i tried before, first pass upscaling with most models starting with high quality videos was not ideal because they would often remove / smooth surface detail, or they would occasionally break in a noticeable way (often in dark scenes, or faces at a certain distance). But if thereâs a workflow that combines another model with certain settings and Nyx XL and doesnât do that, Iâm all ears.
Dumb question, how do I make my license âAuto-Renewalâ? Iâve looked all over and I canât find any mention of âAuto-Renewalâ. I have a subscription, itâs active, I have a payment method in there. But it has never auto-renewed in the past.
Letâs hope they fix it before the release⊠I suppose they bypassed something related to CUDA to make it work with AMD and Mac chips, and in this process SLM lost accuracy and quality⊠a lot !
No, sorry, I am just starting testing Nyx XL right now, as second enhancement combined with Proteus on a noisy 1080i indoor footage. I asked you because the combo was mentioned in the forum, perhaps even in the release notes for 2b, canât remember exactly.
