I have a fully-loaded Mac Studio. There needs to be a way to run the models on my Mac, even if it takes days. Otherwise, the new features are useless. I really need to fix focus misses in 1-hour 4k videos.
You can currently run every model aside from Project Starlight locally provided your system meets or exceeds the System Requirements for Video AI.
Project Starlight is not able to run locally due to the hardware requirements it has at this time.
New features we cannot use (due to cost) do not really help. I have not inproved any of my videos using Video AI. Alas.
1 Like
Starlight is very expensive for private users it is not a replacement. It’s optional, that is ok. What I criticize is there have been no improved local models for a longer time and certain bugs have not been fixed for months and we get new bugs with releases too often, things that worked before. So I aksed myself why I should update.
1 Like
Are you having issues running the existing models on this Mac system to address the issues in the video?
Yes. For out of focus videos. Topaz AI does this nicely for stills. Why cant the same algorithm be applied to a video, frame-by-frame, even if it takes days. But since a frame varies just a bit from the previous frame, there ougbht to be a wwau to speed this up.
Video AI and Photo AI use different models for their processing tasks. They do not share models and will not produce the same results.
There have been some users who have taken their videos and exported them as image sequences and then loaded those images into Photo AI for batch processing and then converted those frames back out to a video file. However, there is not a guarantee that the AI model will retain continuity as it moves through the frames since it processes each frame as a stand alone and does not keep settings consistent when going to the next image file.
You need to figure out a cheaper way… Otherwise, there is no reason to upgrade.
1 Like
Even if the price of Starlight is halved, it will be too expensive for private users who have dozens of videos they want to upscale.
That’s true but maybe this problem can be minimized or solved. Photo AI delivers very sharp Images you can generate additional frames from it without or low “soap effect” doing frame interpolations. Doing high fps excessive differences between frames are gone by transitions.
This becomes more and more relevance: Modern TVs, OLEDs have response times of 0.1ms, tittering is a big problem, and this can be solved doing high fps, this is the future. The Photo AI way should at least be investigated to see what potential the method has and cpu power can be reduced for example by only having to recalculate changes from one frame to the next. Doing Image sequence optimization could be part of photo AI …and then at some point, Photo AI could grow together with TVAI.
