Hello Guys,
As you can see, I'm new on this forum.
And I have a question in my mind since 2 years. That's the question:
How can I remaster Queen's clips, to look like Blue Ray quality ?
I know how to use After Effect, Premiere Pro, Sony Vegas.
If anyone has an idea to make this realisable, it would be very kind.
Thanks for your time and attention.
Have a great day.
1. You already asked this once.
2. You can't really do that.
Study this forum ( link ) if you want to learn and forget all 3 programs you mentioned. They are NLE's, they're for editing not processing.
Actually, there's this whole bunch of super-resolution techniques that do exactly that! I am a geospatial analyst, and these days we use these and other AI methods to extract information from satellite imagery. Super resolution can provide x2 and even x4 boost in definition (think of that as resizing the picture with "simulated detail"). Of course, these algorithms tend to be computationally hard and require a lot of training with suitable datasets.
For video, it might be usually even better as some techniques can use the information present in several adjacent frames (calculating the movement offset) for the enhancement prediction. I've seen some tests for one of this time-domain super-resolution methods and it certainly looks impressive!
Yes, but it's a broad and deep topic. I've tested the Gigapixel AI for video a lot. I made an upscale of DSMN from Sikke's 480p youtube source and 720p Barcelona source to 1080p, they're both shared here and on my channel. Both had to be processed heavily with Avisynth beforehand and upscaling involved at least a few passes where adjustments where made in between those passes also by Avisynth. Needless to say, those were fairly fresh film transfers which plays a huge role. There are MANY variables to create a good video using AI. You can't just take any video out there and expect HD, it will never happen. So I still stand by my statement that one can't do that. Not yet anyway. Closest I got to true 1080p was by taking a 720p film source, but that's not saying a lot. Otherwise for tape sourced videos I can get similar results using only the good old avisynth, they will be sharp but by no means blu ray quality. AI may give better results in certain circumstances, but you need a really damn good source to begin with for it to become noticeable.
It is a deep topic, indeed. Theoretically, if you use a powerful GPU server and you do sufficient training using a custom HD/degraded dataset similar to your footage, most of the AI algorithms would provide very good results - far beyond of what any commercial software offers. But, as you said, you cannot use this in every video and expect good results. For example, I've found that MPEG-2 sourced videos, even if you don't see obvious blocking on them, have lost too much information for the algorithm to perform at its best.
BTW, this is the kind of technical chat I really enjoy, Chief (mostly because I learn a lot). It is such a pity it will be lost here in the Announce section!