It’s rough… but I’m still learning to tweak the settings etc. I used a normal cell recording without preprocessing other than cropping to 512 width with FFMPEG.
Processing Time 7.2 mins on my 3060Ti (0.68sec/frame)
princess jasmine
This is pretty cool. I’ve seen that newer video modifying tools have a better concept of image stability. Seems like the future is going to be wild when this stuff has the quality that individual images do right now.
This is the “fast and dirty” method fro GPU poor folks like me, who want to work locally still. You can eek out a lot more image to image coherence using AnimateDiff.
Is this some extension to automatic1111, or separate project? Can you share a link?
It’s an extension for A1111, my previous post here has more info and links
Can you link the source video for comparison?
Nope, sorry
You’d be better off using an upscaling AI to do this rather than an AI that tries to reinterpret what’s already there. Go grab a pirated copy of Topaz Video Enhance and you’ll see what I mean.
I think the point was to reimagine the existing video, no? Isn’t video enhance just for upscaling, interpolating, etc?
