1
$\begingroup$

(Preamble: this scene is for non-photorealistic rendering; it gets heavily processed in compositing for further use in artwork, so the render not being photo-realistic is fine.)

I've attached a gif below to demonstrate the issue:

Animation flipping between renders with and without shelf lighting

Basically, I have a bartop with some spotlights overhead shining down on it, and on the wall is a shelf containing some bottles. I wanted to add some spotlights on the shelf shining upward, illuminating the bottles, but something weird happens: when I add the shelf lighting, the area behind the shelf gets darker. Not perceptually darker, like an optical illusion—actually measurably darker, as shown in the gif, where the same point is sampled with the shelf lights on and off.

I'm stumped as to what's happening here; in what circumstance does adding lights darken other areas of the environment? There's no environment lighting of any sort; the only lights in the scene are the spotlights. If the lights are extremely bright, without any tonemapping (there's none turned on that I know about) I'd just expect them to blow out the image, not reduce brightness elsewhere in the scene.

Things I've tried to no avail:

  • Turning off clamping altogether
  • Increasing max bounces in the light paths
  • Turning bounces off altogether
  • Turning off color management altogether
  • Rendering with 4x samples (2000 vs 500)
  • Switching from GPU to CPU rendering
  • Switching from 8-bit to 16-bit color depth in Output

Some of these attempts changed the effect, but none of them eliminated it. Oh and, these images were taken before any compositing; this is the render itself. Any idea what's going on here?

The file itself can be found here:

(Note: the compositor has some pretty complex stuff going on but as stated, these images were taken from the unprocessed render). Also, this image is pulled from the "Shading" scene, but it happens on the "color" scene too.

$\endgroup$
7
  • $\begingroup$ Curious. Could you share a representative file, (with no important IP in it, if that's a concern,) on blend-exchange.com ? $\endgroup$
    – Robin Betts
    Feb 2, 2023 at 19:10
  • $\begingroup$ Yup! I've edited the question and added the blend file to it :9 $\endgroup$
    – Arcane
    Feb 2, 2023 at 19:22
  • $\begingroup$ Thanks very much. Somebody else may be able to crack it open before me. Is this of any relevance, do you think? $\endgroup$
    – Robin Betts
    Feb 2, 2023 at 19:31
  • 1
    $\begingroup$ All you need to do in that file to see the effect is show or hide all of the "Booze Spots" in the Lighting collection (they're parented to a convenient Empty to make it easy to find and select them) and render the "Shading" scene. However, if you mean reproduce it in a different file, the materials in that scene are all a standard white Diffuse shader (not Principled—and the little extra in the material is just a tweak to make them a different color depending on the scene), and the spotlights are just ordinary spotlights, nothing crazy. $\endgroup$
    – Arcane
    Feb 2, 2023 at 20:52
  • 1
    $\begingroup$ Great you've got a fix. This is an unusually well-constructed question, and you've been very thorough, and helpful. You deserve it :) $\endgroup$
    – Robin Betts
    Feb 3, 2023 at 5:42

1 Answer 1

1
$\begingroup$

With this minimal reproduction, raising Indirect Light clamping from 3 to 10 (Blender default) seems to get rid of the issue completely. This also shows that the issue is not in the materials or the geometry.

In the full scene, you said you've tried unclamping the values, and it didn't eliminate the problem, that's not what I'm seeing either: comparison. And that's just indirect, you could also unclamp direct and add more bounces.

The top of the spotlight on the right seems to change a bit, but that's a direct light and could probably be solved by unclamping that as well. Seems minor.
Other than that, a couple of bar stools appear to get darker. This seems to be the denoiser's reaction to the increased contrast due to the wall behind the stools getting slightly brighter, and can be confirmed by rendering at higher sample count (both: 2000 samples, 4 bounces, clamp direct 10, clamp indirect 10).

I'm not sure why additional lights would affect contribution of existing lights, but at the end of the day, it makes sense for walls that see no direct light to be affected by clamped bounced contributions.

$\endgroup$
2
  • $\begingroup$ You say "unclamped", but the pictures are labeled "clamp #", suggesting a high clamp ceiling but clamping nonetheless; when I say "unclamped", I meant in the question that I set clamp to 0 (on both direct and indirect), which I understood to mean no clamping at all. I thought that higher clamp values approach 0, and since 0 didn't fix the issue, I didn't bother playing around with high clamping values because I assumed that if the highest clamp ceiling didn't help, any lower clamp ceiling would be even less effective. $\endgroup$
    – Arcane
    Feb 2, 2023 at 23:08
  • $\begingroup$ That said, I couldn't tell you why my Indirect is 3; probably a holdover from some tutorial I followed many years ago and haven't changed because it never was an issue until now. Setting it to 10 does seem to have fixed the issue; marking answer as correct, thank you :9 $\endgroup$
    – Arcane
    Feb 2, 2023 at 23:09

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .