I'm working on a restoration of John Woo's The Killer and I could use some advice.
The source I'm using is fairly noisy and the grain has a fairly unnatural look to it. I'd like to clean that up the best I can.
Here's a clip:
I've been playing around with Neat Video but I worry that it's removing too much detail. I've heard that I could use AviSynth to achieve better results but I'm not sure which plugins and settings would be the most appropriate.
I'm hoping for some help in that regard.
My plan is to make the video look cleaner, then maybe apply a bit of mild sharpening and some color correction using a different source as a reference.
I also have a high-quality scan of 35mm film grain that I could add as an overlay in case the denoising ends up making everything look too smooth and un-film-like.
I'm after the maximum quality possible regardless of render times. I have some familiarity with AviSynth but I'm by no means an expert. I'm absolutely open to complicated scripts if they get the job done.
+ Reply to Thread
Results 1 to 6 of 6
It's not an MPG. Why'd you name it as one? Someone clipped the black levels to 16. Was that you? Has this thing already been reencoded?
I generally don't like denoising. And upscaled SD is harder to deal with. Even when the upscaling is really well done it's obvious it's upscaled SD. Do you find this to be an improvement?
Are you sure it's an upscale though, as opposed to relatively poor true HD? As rough as some of the scenes look, this transfer has a lot more detail than upscaled Hong Kong films I'm used to.
Your clip does look pretty good, better than I've been able to manage with the testing I've done so far. Can you tell me the script and plugins you've been using?
My current plan is to clean up the source a bit, then upscale to 4K, then maybe add some mild sharpening.
I hesitate to ask because I know the tool is not too popular among some people, especially AviSynth gurus, but do you have any opinion of Topaz's Video Enhance AI?
I read some posts on this site from LordSmurf and others really criticizing it but I've found some utility in the tool for certain applications.
More broadly, if I were to upscale a 1080p source to 4K, then apply some sharpening and other editing at 4K, then finally encode it back down to 1080p again would this result in any benefit or would it all be a waste of time?
Here's what I did in AviSynth:
LWLibavVideoSource("TK_test_out.m2ts", cache=false, prefer_hw=2) BilinearResize(960,540) # alternate: Spline16Resize(704,480) SMDegrain(thSAD=300, thSADC=150) aWarpSharp2(depth=5) Sharpen(0.5) nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1920, fheight=1080) ChromaShiftSP(x=-2, y=-1)
After doing some comparisons I can confirm that this is giving me better results than the other denoising methods I had been using. This is especially evident in faces. The other methods, whether using Neat Video or other AviSynth scripts I was playing around with, would leave weird blocking artifacts around faces and sometimes seemed to smooth out details too much in other areas.
This script seems like just the right balance, leaving faces looking clear and artifact free, while retaining details.