Its easy to tell if you look at promotional materials. If this is Anime, lines are MEANT to be sharp. Lack of Sharpness is a resolution artifact. This can legitimately take 10-20 hours just to write a script to clean up a video before rendering and before upscaling.Color profiles should remain the same or upsampled. You need to make sure as little of the original footage is altered as possible.
This is not an easy one button click job. Not trying to make a dollar and drip feed access to decent tools. And they are free and constantly pushing the physical limits of what's possible. These tools far surpass what Adobe and Topaz Labs are able to give you. They are community driven tools based on Python and extremely high end mathematical papers and implementations of edge detection, denosing, anti-alising etc. There are multiple extremely high quality tools that can help you do this. What matters is either starting with a clean source that you want to upscale, or, cleaning up a source.Ĭleaning up a source involves, denoising, dehaloing, deinterlacing/inverse telecine/reverse telecine.It also involves cleaning up block artifacts etc etc. It can not fix your broken video and it will not be able to do so for 10-20 years if ever. That being said, what a lot of the ESRGAN model community like to do however, is create the most pristine models trained on the best images only to use them on the worst footage. The best methods for upscaling using a neural network are ESRGAN based tools such as: Such as fashion, advertising or printing. Because the people using Super Resolution are a niche industry and expect the highest end results. The only acceptation to this is Adobes Super resolution which genuinely is decent for photos. You are paying money to save time i.e drop a video and click 3 buttons, not get the best results. If you are using this you are receiving bottom of the barrel results.Īny company that provides an AI solution is not providing quality.
0 kommentar(er)
