117 points by ux 1 day ago | 6 comments
purplesyringa 2 hours ago
The illustrations were cool. They could've been just prerecorded animations, but I really appreciate making them endless. Quite mesmerizing and motivates reading.
ux 1 hour ago
Hehe, thanks. The thing is, it could hardly have been prerecorded animations :D

I don't know if you tried before, but compressing noise is particularly hard, so 16 videos would have been quite too much of bandwidth. The total size of the shaders is something around 70kB (not minimized) for endless lossless videos, and since I had to write them anyway if I were doing records of them, it was really a no-brainer to embed them to be honest.

dib258 11 hours ago
Thank you for explaining this better! This is still a bit complicated on the Math side but it’s well illustrated to see the result.
Dusksky 17 hours ago
What an impressive and thorough deep dive. I remember first learning about gradient noise through shader examples online, but never realized how much complexity lies beneath the surface. In my own projects, I’ve definitely been guilty of assuming “it looks good enough” without considering all the finer mathematical details. Seeing how derivatives and numerical stability play such a key role really gives me a new appreciation for how much work goes into getting these effects to look smooth and natural. Have any of you ever gotten lost tweaking fade functions to get that perfect wave-like look?
ux 2 hours ago
> What an impressive and thorough deep dive.

Thanks!

> Have any of you ever gotten lost tweaking fade functions to get that perfect wave-like look?

You cannot use anything for the fade function, because you likely want the derivative (and potentially the second derivative) to be 0. See https://gist.github.com/KdotJPG/417d62708c76d53972f006cb906f... for making different ones. I personally never tried anything else than the hermite and the quintic.

nigels-com 17 hours ago
One thing I've been meaning to look into - how to adapt 3D perlin noise to produce gaussian noise - given a specified (scalar) mean and standard deviation.
ahmetrcagil 8 hours ago
Why not generate gaussian noise from scratch? By definition it should be indistinguishable.
nxpnsv 10 hours ago
Pretty and informative!
curtisszmania 18 hours ago
[dead]