How to check the quality of your HDRI's

I’ve just posted a video on my YouTube channel which demonstrates a simple method for checking the strength of any HDRI. Given how popular HDRI’s are for rendering, I’m sure everyone has used them at some stage, but you may not have gotten the results you expected. This is usually because the HDRI is lacking in dynamic range and probably doesn’t have either the full intensity of the sun or the correct colour of the sun.

So I made this video which explains a little bit about the differences between 8 bit, 16 bit and 32 bit files and also illustrates the difference between using a clamped HDRI, (which doesn’t include the full intensity of the sun), and an unclamped HDRI (which does include the full intensity of the sun).

[video]https://youtu.be/kZBAxIOv3t8[/video]

Good explanation, but doesn’t really touch why clamped HDRIs have their use, or how to rescale a clamped to proper values if you need it. I missed some though, because I managed to fall asleep, lol (because I’m exhausted, nothing else :)). The big problem with unclamped HDRIs especially in Cycles, is the amount of noise it produces combined with its difficulty of handling it (currently at least).

See also this Renderman shading movie:

Which also states the sun strength to 2^16’ish rather than 2^15’ish, although a low sun would probably contribute to that.
I have one HDRI that is particularly troublesome. It’s clamped but I scale up the brightness levels for the brightest parts, so it’s not really clamped when using it. The problem is that it has a skyscraper with lots of windows (most likely with an additional reflective coating, and thus a really bright reflection of the sun coming from another point in the image. Two bright “suns” in the image cause insane amounts of noise. Even using HDRI as single source of light, current state of MIS just can’t tackle it.

So for me, although I prefer unclamped, it boils down to how much noise I’m willing to handle. I’ll happily take a fake clamped HDRI and/or bigger sun and/or sunlamp to help out for a 5 minute render compared to a 5 day render using unclamped HDRI only based sun.

Another thing worth mentioning is that of physical based rendering, where “all materials have fresnel and specular” (minimum 4% at F0 for dielectrics, as a general rule). 4% is not a lot. An 8 bit image, or even a clamped HDRI won’t really show up as a reflection in a 4% reflective material. But pump up the sun to correct values and the reflection will show up as quite a nice hotspot. This is what we used to control in old blinn phong based materials and light lists (or custom written to handle it, if the system allowed it). Now we just “rely on HDRIs” and completely fail to recognize that the hotspots are completely lost. And then we call it “more real because it use HDRIs”. It drives me absolutely nuts! :slight_smile:

But yeah, very informative and I’d recommend this to anyone that uses HDRIs.

Hi Carl, thanks for your comments and the video link, it was very interesting and explains the evolution from a traditional rendering approach to a physically based one very well. It pretty much reinforces what I was saying about unclipped HDRI’s albeit with a lot more technical detail.

As regards getting a lot of noise, this hasn’t been my experience, at least with modern rendering software, I test my unclipped HDRI’s with several renderers including Cycles and I can get nice clean renders. There was a time when a HDRI with a real bright cluster of pixels would cause a lot of noise, but I think with modern ray tracers those days are behind, maybe its all down to MIS.
With your skyscraper HDRI you could try selecting the bright windows in Photoshop and bringing down their brightness.

As regards the sun strength being closer to 2^16, yes I’d agree. I tend to use my HDRI’s at an even brighter level than I demonstrated in my video, however as I sell them commercially I’m a bit reluctant to have them quite so bright in case it causes problems for some users, so I tend to err on the side of caution.

Now we just “rely on HDRIs” and completely fail to recognize that the hotspots are completely lost. And then we call it “more real because it use HDRIs”. It drives me absolutely nuts!

This is actually one of the main reasons I made my video, the use of HDRI’s is so prevalent now most people seem to think that just using one will guarantee them realistic results, but given that the vast majority of HDRI’s available are clipped, some quite badly, this is rarely the case.