I'm just once again pointing out that there is a HUGE difference between light and images that are manipulated through a LENS, with its own imaging characteristics, its own lens drawing characteristics, its own bokeh signature, and its own flare/ghosting, and spherical aberration profile, etc.,etc.
Most of the defects in a lens are "linear," in the sense that they are additive or multiplicative types of effects. OR they are technically nonlinear, but in a pretty predictable way, getting stronger with linear distance from the center of the lens image (I guess you could call that "quadratic")
There are a few exceptions. One example being strong mustache distortion (which gets stronger, then weaker, then stronger), but these are indeed exceptions, not the rule. Coma, SA, CA, vignetting, etc. most characterizing defects of a lens pretty much increase predictably and linearly with radius.
Any such effects that have linearly scaling magnitudes like that are trivial to replicate with software. Will your boring old run of the mill photoshop filter make it look EXACTLY the same as XYZ specific lens? No. But it will look like the average effect across various lenses, some of which have barrel distortion, some of which have pincushion distortion, and so forth. AND if you really wanted to mimic a particular lens, you could very well do so by simply scaling the strength of the Gaussian blur algorithm with distance from the center of the image, in the same way that that particular lens scales its defects with distance from the center. For most modern lenses, these scaling factors are known and published (for use in lens correction profiles), and an enterprising programmer could apply the already-measured values to a custom diffusion replication filter with almost no errors.
And even mustache distortion could be incorporated if you cared enough. Few shots of a grid, run it through the right software, figure out the strength*radius curve, and then just apply that to the strength of the Gaussian blur, and you're done.
The point is that there's nothing particularly magical about applying an even, algorithmically predictable effect like diffusion BEFORE a lens versus AFTER a lens, as long as you have the lens on hand to measure and correct for, which we do.
Polarizing filters, by contrast, are ENTIRELY different, because the effect depends on NON-predictably, NON-linear aspects of the specific environment of a specific photo. And thus cannot ever be replicated by software based on a non-polarized photo alone. Nor can something like a true tilt shift effect, because again, it interacts with non-predictable aspects of the WORLD outside the lens (object distances) in a way that would require extra information for software to replicate.
Diffusion does NOT rely on any information about the scene. It is only physics and the lens/diffusion materials themselves, which can be mimicked and then applied equally to any scene shot with that equipment. The two types of effects (diffusion versus other stuff like polarizing) are fundamentally, logically different types of things. One is solvable. One is not.
![[No title]](/data/xfmg/thumbnail/40/40302-79b0636c0b67a1ed65f8ad9e01c690e7.jpg?1734174711)
