Hacker News Reader

Comment by phantasmish | Hacker News Reader

One’s an accurate recording of how a real thing looked.

The other’s fake noise.

One’s a real photo from 1890. The other’s an old-timey instagram filter.

It makes sense that some folks might care about the difference. Like, I love my old family Polaroids. I would not want a scanned version of those to have the “noise” removed for compression’s sake. If that had been done, I’d have limited interest in adding fake noise back to them. By far my favorite version to have would be the originals, without the “noise” smoothed out at all.

Lots of folks have similar feelings about film. Faked grain isn’t what they’re after, at all. It’s practically unrelated to what they’re looking for.

View on HN

> One’s an accurate recording of how a real thing looked.

> The other’s fake noise

But since there is no such thing as the real thing, it could just as well match one of the many real noise patterns in one of the many real things floating around, or a real thing at a different point in time with more/less degradation. And you wouldn't even know the difference, thus...

> It makes sense that some folks might care about the difference

Not really, it doesn't make sense to care about identical noise you can't tell apart. Of course, plenty people care about all kind of nonsense, so that won't stop those folks, but let's not pretentd there is some 'real physics' involved

But… of course there is? A record of a real thing is different from a statistical simulation of it.

I think you missed the "a" vs " the", you can encode different sources that would have different grains, or the same source would have different grain at different times.

But also a simulation called compression of a real thing is different from that real thing, so that purity test had already been failed

I just feed AI the IMDB summary and let it re-create the movie for me. Just as “pure” as high-bitrate h.265, after all.

You've chosen your argumentative perch very well, it's indeed right down there with the AI slop where you can't see any difference in reality

Well film grain doesn't matter because compression exists, apparently, and may as well be simulated because it's already failed the "purity test" and may as well be algo-noise. That holds for everything else! May as well maximize the compression and simulate all of it then.

[EDIT] My point is "film grain's not more-real than algo noise" is simply not true, at all. An attempt to represent something with fidelity is not the same thing as giving up and faking it entirely based on a guess with zero connection to the real thing—its being a representation and not the actual-real-thing doesn't render it equally as "impure" as a noise-adding filter.

You're still dancing there in the slop, hallucinating the arguments thinking it's a pretty dress!

It may as well be stimulated because you won't see the difference! So now you've imagined some purity test which was never true, so you have nothing and start hallucinating some hyperbolic AI thing

> But also a simulation called compression of a real thing is different from that real thing, so that purity test had already been failed

Quoted: the introduction of “purity test” to the conversation, from not one of my posts.

Reply on news.ycombinator.com