by ffsm8 2 hours ago
Upscaling and even de-noising is something very different to applying filters to increase size of lips/eyes...
It's a different diagnosis, but the problem is still, "you transformed my content in a way that changes my appearance and undermines my credibility." The distinction is worth discussing but the people levying the criticism aren't wrong.
Perhaps a useful analogy is "breaking userspace." It's important to correctly diagnose a bug breaking userspace to ship a fix. But it's a bug if its a change that breaks userspace workflows, full stop. Whether it met the letter of some specification and is "correct" in that sense doesn't matter.
If you change someone's appearance in your post processing to the point it looks like they've applied a filter, your post processing is functionally a filter. Whether you intended it that way doesn't change that.
Well, this was the original claim: > If his content was about lip makeup they make his lips enormous and if it was about eye makeup the filters make his eyes gigantic. They're having AI detecting the type of content and automatically applying filters.
No need to downplay it.
I didn't downplay it, I just wasn't talking about that at all. The video I was talking about didn't make that claim, and I wasn't responding to the comment which did. I don't see any evidence for that claim though. I would agree the most likely hypothesis is some kind of compression pipeline with an upsampling stage or similar.
ETA: I rewatched the video to the end, and I do see that they pose the question about whether it is targeted at certain content at the very end of the video. I had missed that, and I don't think that's what's happening.