Hopefully Starlight Precise 2 will one day if the boffins can find a way to make it work with current GPUs but, at this present time it’s restricted on the Cloud
1 Like
Apollo frame interpolation always prompt ‘recoverable error’ with my RTX5090, lower the max memory to 95% solve the issue although it only occupied 17% out of total 32GB VRAM in 100% allocation, weird. I recall similar bug inherited in different AI model randomly very long time ago even when i using GTX1080 in Video AI 2.x and RTX3080Ti in all the subsequent releases.
I have the same problem with my 5090. Also, Chronos on my 5090 for some reason doesn’t work properly…it doesn’t convert the videos to 60FPS even though it claims to be doing so. The only frame interpolation model that works well for me is Aion but that one is unfortunately very slow compared to the others.
I tried both Chronos and Apollo fast, both works fine with my RTX5090 with 100% memory allocation, only Apollo need to lower down to 95%. Anyway i’m using Nvidia studio driver, not game ready driver.
I have this ‘recoverable error’ also several times, happened when I was using Aion frame Interpolation. Overal my impression is V1.1 is not very stable, for example using intense viewslider in the output window could kill the rendering, then error appears something like “user action aborted”
Is a 5060Ti 16GB good for Topaz Video? Its the bottom rung for 16GB GDDR7 RAM.
Can pick these up for a little over £400
It depends on your definition of “good” and what models you plan on using. If you plan on using Starlight Mini or Sharp locally…then IMO the results won’t be great unless the videos you’re upscaling are simply a few seconds to a minute in length.
If you plan on using Proteus or Iris, you would have fairly decent results at 1080p.
For me, I needed to build a new rig with a 5090 since I prefer to use Starlight Mini for most of my upscaling.
Its the cost £2500!
Im looking to do Starlight Mini
Would a 5070ti be a better option?
Exact same hardware specs as you, exact same experience.
16GB Card ist acceptable for Starlight, when you do SD footage min. or 2x upscale. If you plan doing up to 4k projects this isn’t a viable option, as higher resolutions also require more VRAM, and the difference compared to a 32GB card becomes very noticeable. Besides, this would be unusably slow with a 5060.
I do 4K stuff with a 5090 and sometimes with TVAI 7.0.0.4.b. there the most extreme I’ve seen was more than 48GB GPU RAM usage, 31.5GB VRAM, and the rest shared. Rendering speed 0.1fps or less 
I use a 5060Ti with Starlight Mini and Sharp - I prefer mini. I upscale to HD and not 4K. It has worked fine for me. It is slow and I wish I could afford something more expensive (and the electricity bills that go with it) but it does the job.
I am generally upscaling SD files (768x576)x3 with Starlight Mini and then I do a second pass with Iris to change them to 1920x1440 which I can then pan and scan in Resolve to make a widescreen image.
I also have access to a system with a 4090 which is roughly twice the speed.
This is a valid flow. I do something different but can’t say if it’s better. I do the width stretch 768x576 to 1024x576 into Hybrid or Virtualdub (and deinterlace) then I give Topaz this. But sure upscale time with Starlight is higher because there are more pixels.
There are many ways. people who have no highend card can put left/right letterbox arround 4:3 content to generate a 16:9 source, then Starlight does 720 mimimum upscale instead 960. Then with the 720 you can do a 2x SLS run, ends in 1440 or do with the 720 footage 2x Iris/Proteus upscale, gives you again 1440
OR scale the 576 source with lanczos (or bicubic Spline 2 from Hybrid i prefer) down a little bit to 540 and do SLM 2x and you get 1080, afterwards if you like run this trough Starlight Sharp doing 1x and the SLS model does not use NYX because it’s 1080, often endresult looks pretty good because the SLM->SLS combo works.
I’ve got one and found it reasonable, but I have it in a secondary system that I can just run 24/7 if need be for workflows. I wouldn’t suggest one for your primary system unless it’s just for short(ish) SD to FHD video upscale work.
There is no point in increasing it by 3 times and wasting time, optimally by 2 (there is no difference), then either Nix or Rhei, you can smooth out AviSynth++ if you have enough sense.
Doing Starlight 3x instead 2x gives finer results, but instead doing 3x SLM, you can also invest this time into doing SLM 2x and additional running SLS 1x (as finsiher, no NYX happens) or do SLS 2x (NYX happens) over the SLM footage (gives you 4k). In my opinion valid options.
There’s no difference, take your video and let someone else process it in 2x and 3x, and let you watch it, you won’t be able to tell the difference…
???
There is a MAJOR difference in quality between Starlight 2x and 3x - especially for text, graphics and logos.
Maybe not really “major”, but what you see in faces also applies to everything else, it’s just that you see it best in faces because humans are trained on how faces looks. So yes 3x resolves details a little finer than 2x, but also requires more VRAM, so I wouldn’t do 3x with a 12GB card.
Topaz 1.04 is the last version that I can use the Preview. The past 2 updates return an Error when trying to preview settings.
