That is some serious dedication. I’m waiting a bit before processing any long videos like that, hoping there will be a speed increase coming for starlight mini, or a better starlight model that is faster with better quality. It will come, it just depends on if we are talking weeks/months/years.
Kyle, is there a technical reason why Astra is only accessible by web and not through the Topaz Video UI, or was it purely a marketing decision?
The original design and purpose of Astra was for a different workflow with content generated from other AI sources. As development continues on new features, tools and models we will see what decisions are made about the future of Astra and Topaz Video UI for cross sharing of model access.
Just some feedback. I’d love to be able to use Astra and those types of “ai to real” models in Topaz Video. That will keep me subscribed for a very long time if I can use those models locally.
If you have the full Topaz Studio bundle subscription, you already have access to Astra with your account, just need to use cloud credits for processing just like the cloud workflow from the desktop app. You can launch astra from the Studio Account Page.
Starlight Precise 2 → Cloud only
Nyx High-Fidelity → Cloud only
Nyx XL → local but Ai is 1x only, while Topaz use NYX XL in Starlight Sharp for scaling to any resolution. why we get just 1x?
Proteus Natural → That’s really useful, thanks.
Batch Render, Enterprise API → okey, the killed the CLI
GPU Optimizations → not faster for Windows and some bug fixes for Apple, but unfortunately still not good
Unified Interface → yes the interface, everybody hates
But I have one more positive point: New Grain, gaussian is good. However, it could be even better if grain were combined with Gaia as second model, what Gaia does with grain has potential. Unfortunately, add grain cannot be applied without selecting a model.
And we get Starlight Sharp, can be useful pends on..but it’s FHD model only and when you uspcale you can’t select FHD as target resolution, this is strange.
What bothers me most is that there is no progress with SLM and that upscaling is limited to fixed factors of min, 2x, 3x, and 4x.
I agree with you.
I agree. It looks so much more natural, if you want to get rid of the “plastic” artifacts generated by a particular model.
Hi, well, I was doing the rendering when I wasn’t at home. I had cut the video into several pieces of different lengths. The lengths varied depending on the cut transitions. It was a family video from 1991 on DV tape, which was then transferred to mini DV before being converted to a digital file format. So I calculated which segment I would render based on the amount of time I would be away from my computer. Pretty handy.
Batch render is not a functional equivalent of the CLI; it’s not even close.
The last update was back on January 23. Is there any chance we’ll see any of the following happen sometime soon, at least within this year?
-
Better speed and/or quality for Starlight Mini on Nvidia GPUs
-
Starlight Sharp being able to upscale past 1080p natively instead of relying on Nyx XL
-
Local versions of Starlight Precise 2 and/or Starlight Creative
It’s been almost a year since Starlight Mini launched with Video AI 7, and I haven’t really noticed any improvements in speed or quality for that model since then.
way too expensive and you lose privacy
Please do not wish for faster processing for Starlight Mini. That would lead to worse quality most likely… I am very happy with what it is now. The processing is running 100% on the GPU.
The last 7.x versions were buggy. For example, Star Light Mini is unusable because it can no longer be downloaded/installed (endless loop). Support? None.
We’re now paying for version 1.x.x via subscription, but we’re not getting any continuous development. I feel stupid and ripped off.
Red card from me.
Red card from me.
Seems there is something wrong with your install man…
Mine is working just fine. Windows 10. Try installing and running it as administrator?
Of course, I just meant better speed, but only if we can get it without sacrificing quality.
I’m using 7.0.2 all the time here…
Yes, this absolutely. I don’t want worse quality. I want faster speed. It must be possible with some kind of optimizations. Nvidia is always releasing new cuda versions and other things like that.
Thanks but I already have a strong GPU. I don’t want to use credits and I don’t like using the cloud for anything because of privacy reasons. Unfortunately for me, it’s as if the cloud offerings don’t exist at all. If it’s not in local, I don’t use it.
I’m still only getting h.264 exports when I have Prores 422 HQ selected with Starlight Sharp on MacBook Pro M3 Max fully loaded. Why is this?