WeirdShadingBug.blend (1.1 MB)
See attached blend. It renders out as shown below, which is completely absurd to me. To my knowledge, this didn’t happen in Cycles until I made a walkthrough animation in Eevee, where the bug didn’t appear, and reverted to Cycles because I accidentally deleted some frames that wasn’t part of a low quality animation. I’ve tried switching to Eevee and deleting lighting cache’s, but doesn’t appear to do anything. Light groups have been deleted, world have been erased, post processing have been erased. Everything except camera, four walls, and two lighting strips have been deleted and purged. To no avail. What on earth is going on?
I rebuilt the scene in a new file point by point. The geometry is exactly the same, the pivots are exactly the same, the camera is exactly the same and I get similar but different shading errors.
I never switched to EEVEE or anything in the new file so I guess it has nothing to do with that. I’d say it is a problem the emission shader and very long lights.
This does not happen if you replace the emission lights with area lights of the same proportions.
The problem also goes away if you give the plane a couple of segments. The length of the polygon seems to be the problem.
I maybe off the mark here, but that somewhat doesn’t surprise me. Its like the bad pink light is acting as if it only has 1 “light” (likely wrong term, but lets go with it) along the whole length, resulting it harder edges/bands.
While the green light, being subdivided, so its acting as if its a group of “lights” and the results all get averaged out and hence smooth gradient.
The fact you say that an area light acts the same as the green light, somewhat confirms this. As it was my understanding that an area light does in fact act like a group of “lights”, so it all gets averaged and smoothed out.
It is of course very possible I’m totally wrong, not really dived into lighting yet.
Thanks. Wait, how are others allowed to mark as solved? Weird.
Splitting them up into shorter segments (adding loopcuts) did help, but I’m weirded out how badly long striped emission geometry gets sampled compared to long striped area lights. I’ve used similar before without any issues, although often indirect cove lighting or supplemented with other interior lights or sun & sky light - this room doesn’t have any windows though.
There was no ceiling or floor in this one, in order to isolate the problem.
Yeah, I used to do nearly all my lighting with emissive planes and never encountered this problem. I also use these types of long thin lights frequently for all kinds of things.
Perhaps it is just a weird edge case but it’s good to know.
Perhaps this was introduced in the cyclesX refactor? That touched a lot of the sampling code.
What more do you need for this to be solved? I mean, it’d be great if cycles sampled long polygons properly, but in the interim, doesn’t subdividing those triangles solve your problem?
Hehe, I just got jumped last time when I started typing - “already solved” - whuut? So I clicked solved myself - which joseph removed - so I added it right back, lol. Yeah it solves my problem, but its a weird one to run into, so I thought I’d report it as a bug for the devs to look into (or deem “working as intended”)
I’m ashamed for not even trying to divide it up, but in my mindset I was sure there was some previously baked Eevee lighting that messed everything up.
