AI upscaling is still just upscaling. It’s not going to change shadow details like the way this article suggests. And if you upload SDTV video to YouTube, you are ensuring that upscaling has to happen on every display you can watch youtube on. And if you have to upscale, more intelligent, content-aware scalers are likely to be better than older scaler algorithms. Even if YouTube didn’t do this, a scaler on the playback device will.
No sensible engineer is going to design an upscaler that intentionally makes video look bad. This youtuber could probably solve their problem by scaling the video to 4k themsekves, ensuring it still looks shitty after scaling, then upload that version to YouTube.
Of all the uses of AI we might complain about, I don’t really get the beef here.
AI upscaling is not “just upscaling”. Upscaling doesn’t introduce new data other than averaging pixels. AI upscaling will fill in missing details with whatever it hallucinates.
Bilinear resizing is not the same thing as “content-aware” anything, and you know it.
This is happening to videos uploaded in perfectly reasonable quality, and it’s visible when people watch it at any resolution.
They’re just using AI to fuck with video, and not asking permission, and not telling anyone.