I think we all want to run it (starlight precise 2) locally - if so, when?
2 Likes
Thank you Topaz for updating Founder access. I am so glad that I cemented my founder status.
. Topaz products are the best in the upscaling space and I am glad I spent my money with you. Game changing!!
Now a couple of gripes if I may. I have not tested the models yet and look forward to it. But I have a VERY, VERY frustrating sequence of events here that happens way too often. I find myself having to download the Starlight models ever so often. I am pretty sure that the models do not change as often as I use them but it is like the program forgets that I downloaded them five times in the last ten starts. May I be bold enough to say this is just not okay with me. It happens on all of my computers as well so I am certain that it is not a system level issue just on my side. I do not have the time or bandwidth to keep downloading the models all the time. This has been going on for months now. What gives? Is this some way for Topaz to protect the models? I am a little mystified here. It is just the Starlight stuff. I have the models hosted on a different drive rather than my C drive but that program knows where they are.
In addition - the infamous login issues. It appears like this has been sorted on the Astra and Bloom sites but it was just as bad there for a while. But it gets so annoying having to login to Gigapixel and Topaz Photo locally just about as often as I use the programs - it is super frustrating. Again - what gives?
Are topaz programs forgetful - for state of the art I would love for you to remember that I downloaded about 10 Gigabytes of models last week and also that I logged in just a short while ago. This is going to get worse as the Starlight modles increase in size. This does not happen with any of my other professional tools I use. Login, download Gigabytes of data, login again, download gigabytes of data. Rinse repeat, rinse repeat, rinse repeat. Come on guys, and I say this with the greatest of respect - but please address this. I love your products.
Does it download different models based on resolution? Perhaps that is what I am seeing. Now if there is something wrong on my side, I would love to get to the bottom of this. But if not - it needs to be better for us end users. And also if I am the only one on the planet experiencing this then I am happy to trouble shoot, but if others experience this too then I would love a response from Topaz. Peace and keep up the amazing work!!
6 Likes
Informative threadā¦.
Againā¦.kind of the more things changeā¦..the more they stay the same.
For localā¦.for the vast majority of simple minded users like me that do not want to experiment endlesslyā¦..
For old VHS and converted 16mm and 8mm filmā¦
Just bite the bulletā¦.use Starlight miniā¦go on vacation for a weekā¦then use Proteus or Iris on a second pass and violaāā¦itās actually quite good and FAR better than the original.
I also am heartened by Topazās efforts to ākeep pushing the envelopeāā¦..and they DO seem to stand by their promisesā¦.albeit sā¦lā¦oā¦w.
In the meantime, we have a decent āsolution.ā Starlight mini worksā¦quite well. Problem is, I just might not live long enough to get all the rendering I needā¦done.
7 Likes
Most of what is running on Astra (generative AI models) is there due to hardware requirements, but we will see what the devs and research team are able to work out in the future. No promises, though, as there are a lot of variables and the coding of the app is above my pay grade and expertise.
6 Likes
Since Topaz software is geek orientated software, many people have frankestein/professional cards with hight vram 48/96 GB VRAM
We need it, pls make it happen in beta <3
2 Likes
I wish Jensen wasnāt so greedy. Iād pay a reasonable price for a card with more VRAM. Video cards would be far more useful if the VRAM was bumped up considerably.
2 Likes
Will founders on Mac ever get to access SL Fast and HQ? Apple users have been marginalised for a long time, so disappointed to know that these are only for Nvidia.
8 Likes
Itās crazy that Topaz doesnāt give us an FHD resolution option for SLS. The model always outputs 1080p, and the scaling to the target resolution is handled by NYX. Thatās the part I donāt likeāNYX not model-enhanced, itās oversharpened.
But how can you force SLS to output 1080p without NYX? The trick is that the NYX step happens right at the very end. Just before the rendering finishes, copy the unfinished output file. You should have selected FFV1 as the format beforehand.
At this point, you wonāt be able to play the copied FFV1 file because the container header hasnāt been fully written yet. Simply remux the MKV with MKVToolNix, and then it will play.
I know this method isnāt handy, but if your target resolution is 1080p, it can deliver excellent results. For example, if your source is 480p, upscale it with SLM 2Ć gives you 960p. Copy the rendering file before it finishesāyouāll get a wonderful with two models enhanced 1080p output 
and you have both also the finished higher res. version with nyx, sometimes pending on footage it can be used, maybe you can do overlay in video editor (advanced users)
1 Like
When is any of these starlight models going to be publicly available for any nomral topaz labs account using an AMD GPU.
Been waiting a seriously long time and I am glad I cancelled subscription for this year. Iām probably not coming back until this is rectified which I have a funny feeling it wonāt until next gen AMD gpusā¦
Why would I want to be classed asa founder if I still canāt use these founder features?
If these models are so bad to run on AMD gpu then just tell us with a straight face that AMD gpus are not going to be supported on future diffusion models. Thats all you gotta sayā¦
2 Likes
If you upscale the source file to 540p, you will get 1080p SLS output without NYX. 
Yes but choosing 2x or 3x you get at least 960p, then you must do lossy downscale to 540p. I try to avoid downscaling. When you do SLM 2x with 480p source you get 960 and from there you can go to 1080 with SLS, which is model enhanced upscale step without downscale. In short it would be nice if we could get native FHD SLS, since it already does it in background, this would offer more flexibility.
2 Likes
This sounds interesting. Do you have any image comparisons of SLS with and without nyx?
so no linux release? that has been discarded?
3 Likes
yesā¦.
i think you know very well AMD support in diffusion upscaling is so limited outside thatās why you turn your beg to Topaz. To be honest this is not Topaz fault but the whole industry is lean towards Nvidia CUDA due to its mature eco system long way back. Why not you change to Nvidia GPU if you really want diffusion model so badly instead of waiting something that is not promised elsewhere?
I already understood what you have just explained to me a while ago. However I am trying to get a very good point across topazā¦
If they claim that AMD gpu and so forth are supported in there software, then have the balls to say that AMD gpus are not going to be supported for diffusion based upscalingā¦
Thatās all they have to say. I donāt care if I get to use starlight or not using AMD, the point I am trying to make is just tell me as a customer and others as well if AMD RX 7000 OR RX 9000 series are not supported for these models. It clears confusion because the beta users using AMD cards do have access to the diffusion model if I remember.
All of that makes it forever confusing if beta users can but founders or pro supportes canāt.
I use to own a rtx 4070 super. Good card but was one of the worse cards I have ever purchased brand newā¦ never will I do that again with nvidia.
Please, do not support topazās poor behaviour over the past year if not longer.
1 Like
I can do it later; at the moment, both of my PCs are busy.
Beta users have access, because that is the point of Beta users. To check if models and stuff work. And they might have figured out that those models have issues running on AMD cards, making them not ready for releaseā¦
While iām not sure what happen to your RTX 4070 Super, i have RTX3090 and RTX3080Ti dual GPU setup on a AMD R9 5900XT rig, and another RTX5090 with AMD R9 9950X3D rig, none of them giving me any issue so far. I can even run Starlight Mini concurrently (dual instances) on the dual GPU rig without any issue. In fact i started using Topaz AI since it first came out, with GTX1080. Anyway iām strictly talking about Nvidia card on Topaz diffusion upscaling, while you may have unpleasant experience in game, i donāt know. But what i want to point out is, Nvidia card is far better in AI workload in terms of usability and support since GTX1000 series, thatās the fact.
1 Like
Beta users who are running AMD cards do not have access to run the additional Starlight models as they are not currently supported for that hardware. The same goes for Mac systems. They can only run the Starlight Mini model.
1 Like
