If you use MPV you can likely watch the ongoing processing via that, at least if exporting to FFV1 or ProRes… Previews of Starlight never really worked for me (that I know of), so I just do it that way to check it out while it is running…
Topaz video projects are saved in the Documents folder of the user. So, instead of the whole nine yards with a registry cleanup, you can just delete “Default” folder in the “%username%\Documents\Topaz Video Projects”.
2 Likes
I wish there was a model that combined the best parts of starlight mini and starlight sharp. Starlight mini tends to produce more accurate or convincing upscales in my opinion, but I really do prefer the sharpness of starlight sharp. It’s just that sharp sometimes doesn’t look 100% right or the details don’t look as good as mini. For example sometimes upscaled text that was already clear enough to read, has weird artifacts when upscaled. An “A” might have some weird looking pixels around it even though it clearly is an A. I notice that with the lettering, but it also happens in other objects. A few updates ago someone posted a comparison with a bike wheel and in starlight mini you could see the spokes, they were just slightly blurry, but in sharp it looked really weird.
I’ve tried upscaling starlight mini with iris or proteus, but faces upscaled that way generally don’t look better to me than a starlight sharp first pass. Has anyone tried upscaling with starlight mini, then doing a 1x pass with starlight sharp? Is that possible?
3 Likes
Not sure if it is bug or good thing, or Topaz can boost speed with this, I have noticed, in Topaz Video 1.2.1 , on M3 Max 128Gb, if I pause the processing the video in the middle or so, then after resuming the video, I get more FPS, for eg, I normally get with 1080p video almost 15 - 20 FPS but they can go as high as 112 or sometimes even more, here is one screen shot. Video was 1080p, Proteus 4, HDR on. I have had this result consistent with many videos, when video has processed more than 50% or so, then you pause and resume gets more frame rates.
Did it really take only 10s to finish? Maybe the counter is wrong because it only started a few seconds ago and it’s already halfway through the video. I think it calculates the speed by how long it took to get where it’s at and how many frames it still has to process.
1 Like
Yes, it took 10s to finish. It was started a while ago, at resuming the video, it is very high FPS like 5640 FPS or even higher number, I believe it starts from scratch but goes through already ran frames, then it became slow and slow and this was final few minutes with same FPS.
No starlight sharp rendering view is not working. I think it has something to do with post processing, at the end of the main rendering, NYX runs over it and perhaps some other processes as well.
yeah sometimes I do SLM 2x and afterwards SLS 2x with it, then you profit from both, but you end with 480p source at 1920p and from there I do upscale to 2160p. 1080p is plenty for me—but the software forces me to go higher which i don’t like. However, if you use SLS + SLM, often you get halo effects and I must apply a dehalo process, and sometimes also “add blur,” to get the final result.
>Thank you for your continued use of Topaz Video!
Has the bug that has made this thing unusable for the past six months been fixed on this version? Before you ask “what bug”, I’ve been asking the same question with every release and I’ve always gotten a “oh please, we’re aware of this, but we can’t fix it”-answers. If it hasn’t been yet fixed, who I can contact to pay me back for the months that Topaz Video has been unusable for me?
Once again, I don’t want to install the latest patch only to end up having to uninstall it and waste time having to reinstall the older version back.
Here is real sequence, with improved FPS:
after pausing and then resuming the video:
after few minutes:
you see FPS went up from initial 17.9 to almost 50 when it was finishing.
The workflow with Starlight Mini is as follows and is listed on the docs page.
To ensure the best quality outputs, Starlight exports will have a minimum resolution of 1280x720.
-
If your footage output settings are under 1280 width, then the full resolution will move up to match the 1280 width.
-
If your footage output settings are under 720 height, then the full resolution will move up to match the minimum 720 height.
If your source file is of a lower resolution, then it will be upscaled to match this option before export.
What size is the original file that you are working with?
If all of the variables are showing as 720, then that file resolution is very small, and running a second pass to achieve the resolution you are looking to work with will be required.
It has been requested to the devs and research team to work out a way to allow other upscaling options but that has not been worked out yet without reducing the quality of the models output.
2 Likes
Starlight sharp appears to have problems recreating convincing looking teeth. Have you guys noticed any other downsides? Text is also not great, as I mentioned above. Eyes look pretty good though and much better than starlight mini.
Running a starlight mini pass first and then starlight sharp at 1x after that kind of oversharpens the overall video, but it also ruins details like teeth.
1 Like
SLS seems to invent more new things that does not exist than SLM (like textures, teeth, etc.). To me, it feels like the effect is weaker when using SLS upscale instead of 1×, but I’m not completely sure about that. SLS 1× definitely has the jagged outline problem, similar to Rhea 1× and this could be very worse so I don’t use SLS 1x.
What works well is SLM 2× first then SLS 2×. Sure can still produce artifacts, but it also has the advantage of good denoising and clear, sharp results without jagged outlines. However SLM+SLS Upscale tends to create oversharpen and halo effects, so using dehalo or/and adding a bit of blur afterwards helps.
1 Like
Awesome, thanks for the tips. I generally like to use SL mini at 3x but that’s if I’m not going to do any further upscaling. I will test SL mini at 2x and then upscale that using SL sharp at 2x. Strange that SL sharp at 1x produces more weirdness than at 2x.
