oh no, this is bad news…
Any info on when the next update could arrive? It’s been more than a month since the last update.
Rhea1 takes a lot longer to process than Iris and I am wondering if the results are enough different to justify the much longer processing time, I am assuming you have tried both to compare?
Beta 2? I only see the one for 1.7.0.0b
I think he means macOS by that.
Yep. macOS 27 beta2.. I replied to @domdom mentioning beta3 was acting up when switching off monitor
It’s not like Starlight vs Iris difference in speed.
I use Starlight as a first step, and Iris/Proteus Natural and Rhea as a second step in my workflows.
Both models don’t produce perfect output, so I have to combine them.
Iris or Proteus Natural for a base layer in NLE, Rhea for accents. I place Rhea result over the Iris or Proteus with Rhea opacity about 50..65%.
I push Iris/Proteus layer sharpness to my taste.
A mix of both models produces textured output (by Iris/Proteus) and fine details by Rhea.
Also, I don’t like Topaz grain, so I add grain to my taste afterwards.
It’s tricky to estimate what time it will take to complete an enhancement with SLP. There are 3 phases in the process (stable diffusion model):
1.Encoding
2.Enhancing
3.Decoding
Those 3 phases run on batches of frames (~100) and timing varies a lot for each phase.
What I do in my script is I calculate the time difference between last and first frame (I generate individual frames as TIFF files), so I get a more accurate (not perfect) value.
Hope Topaz will implement a better algorithm to predict remaining time soon.
Just coming across this one now. Wow! That’s a good workaround.
SLP is still magic for me, but after so much extensive usage, I’m starting to notice this ghosting/fast motion blurring more frequently. Going to try out this trick.
I do have a few questions regarding this method though:
- Can we get even better results by using an Apollo/Aion (Frame Interpolation) pass before the SLP upscale?
- What software did you use to increase the framerate in this example?
Yes, you can get much better results when you have a 480p (p=progressive) video and want to double the number of frames per second. To do this for a 480p video, I always use FrameGUI. If the video is 480i (i=interlaced), I always use AviSynth QTGMC. In my opinion, it gives the best results.
Here is quick summary of Starlight Precise performance on 3 systems I have access to. Note that these machines are sometimes doing administrative tasks, hoping they do not much interfere with the performance.
I did the math with 360p videos enhancing to 960p
Min/day spf system
12 4.8 Mac Studio M4 Max 14 CPU 32 GPU 36GB RAM
5.76 10 MacBook Pro M5 10 CPU 10 GPU 24GB RAM
3.2 18 Mac Mini M4 10 CPU 10 GPU 16GB RAM
For 480p videos enhancing to 960p
Min/day spf system
9.6 6 Mac Studio M4 Max 14 CPU 32 GPU 36GB RAM
Love SLP, it needs to go faster, hope this is possible!
Edit: MacBook needs to remain open for better cooling performance, I noticed spf was much higher when used closed
I do simple repeating the frames with ffmpeg. I think Interpolating the frames can also improve the image quality a bit, but it will never be as clean after you convert it back to the original framerate.
To convert a video with 14.985fps to 59,94fps use this:
ffmpeg.exe -i Y:\video\input_14.985fps.mov -filter:v fps=60000/1001 -vcodec ffv1 Y:\video\output_59.94fps.mov
after using SLP I convert it back to 14.985fps:
ffmpeg.exe -i Y:\video\output_59.94fps_SLP.mov -filter:v fps=15000/1001 -vcodec ffv1 Y:\video\output_14.985fps_SLP.mov
at last you can use any frame interpolation software as you like if you want higher framerate.
I’m using RIFE
you can also try to use 29,97fps with 30000/1001
23,976fps with 24000/1001
or if your source has exact framerate like 30fps use -filter:v fps=60 to doubble the framerate.
would be nice to show us the result with double-framerate vs interpolation like apolo
I would not messup with interpolation if you want only simple upscale.
But the best would be to test at first and compare.
Thanks for the suggestions. Will try them out for sure.
Thanks for the detailed reply. Using ffmpeg to increase framerate would be super handy. Will try this out.
Same thing with a Mac
Could you briefly walk me through your use of ffmpeg as prep for Topaz? With an example. Don’t want to take up too much of your time though. Thanks.
Yes, stay away from macOS 27 Beta 3! It looks like any change in audio causes Video AI to crash. I run the app using nothing but Screen Sharing, so I am not connecting any new monitor that could cause audio device to change. But if I approach the mac with headphones, I am fairly sure it switches the connection to the mac and crashes Video AI!
I have been sending logs and crash reports to Topaz and we don’t think there is anything wrong with Video AI. it’s just beta 3 that does something weird.
macOS 27 Beta 2 is OK, dying to get beta 4 with a fix from Apple ![]()
Depends which you use. My studios and mini stay quiet and cool (max 60c).. no problems with heat here..