Industry News

By Jeff Mottle

Effects of HDR vs MDR on Image Based Lighting

When attempting photo-real CG renderings using high dynamic range images (HDRI) with image-based lighting (IBL), you need an accurately photographed environment that contains the full dynamic range available. One of the main benefits to using IBL is the 'drag and drop' nature of it - you load your environment HDRI and render, almost instantly giving you highly realistic lighting.

Before carrying on about this topic, (full disclosure) I'm a photographer of HDRIs at Hyperfocal Design - however this applies to any and all HDRIs. 

Is it really HDRI?

Unfortunately, some HDRIs are often barely higher in terms of range than an LDR (Low Dynamic Range Image/8 bit depth/Jpgs etc). Even if the dynamic range is good, but not great (you could call them "Medium Dynamic Range Images" or MDRI). You can end up with bad colour casting, as your renderer will grab some dull, white light from the tiny sun, plus loads of bright blue sky, and apply that to your scene's lighting in an unbalanced and unrealistic manner.

So while any image can be saved in .exr or .hdri format in 32bit, there will not always be any more additional information than an 8 or 16bit image - it has to be photographed correctly in the first place. Some HDRIs are little more than a 3 bracket spherical environment, so the sun is hardly much brighter than the surroundings. This isn't always bad, as you can substitute the missing sun with an artificial directional light, however you'll miss out on the complexities in reflections, lighting and shadows that you'd otherwise get with a full HDRI. Also notably, a full HDRI can be clamped down if you'd like more control from an artificial light source, but the reverse is impossible.

What's the visual difference then?

What exactly is the difference between a poor HDRI (or so called MDRI) and a good HDRI? Here's some renders!

The first HDRI I've used was photographed with what I'd assume is only a few brackets, and therefore a very clamped sun - a good example of an MDRI. I've blurred the image background to hide any clue to the creator. Underneath it is another HDRI using an accurately captured sun - one of the most accurate ever in fact, by Paul Debevec. http://gl.ict.usc.edu/Data/skyprobes/

Image based lighting render using an "MDR" image environment:

"Medium dynamic range or MDR", rendered with Blender Octane Edition

Image based lighting render using an HDRI by Paul Debevec:

Perfectly captured HDRI sky, rendered with Blender Octane Edition

Light colour

The colour cast I talked about is incredibly obvious in the MDR version, as the environment in the lower dynamic range HDRI creates an awful blue cast - this is essentially unusable without an artificial sun to fill in the range and bring the lighting colour back to normal. Most likely you will need to tone down the HDR strength to reduce colour casting then relight.

Sun strength

When the sun is clamped as it is in a MDRI, and not photographed correctly, the sun is a similar strength to the surrounding environment, meaning you get little directional lighting. It also affects bounce light, which can reflect off of diffuse or reflective surfaces (not easily visible above because the sun is fairly overhead, without any highly reflective surfaces). Here's a render using my own Hyperfocal sunset HDRI:

 

Strong HDR sunlight creates bounce light (caustics) on the ground plane and nearby spheres. Rendered with Blender Octane Edition.

Highlights and reflections

The other fairly obvious downside is the loss of the highlight on the diffuse reflective black sphere on the MDR render. These highlights are lost in any non-mirrored reflective materials, and will also be lost when viewed through tinted glass, water, smoke or fog. If using an animation, these highlights will also blur out to nothing if using motion blur when compared to a real HDRI. Motion blur has been boosted greatly for illustrative purposes:

MDR, blurred. The sun fades out noticeably.

 

A high dynamic range sky, motion blurred to the same level as the above MDR. The sun remains bright.

 

Shadow quality

The last thing lacking from a low range HDRI is the shadows. Because there is so little range, there is no difference between the intensity of light hitting objects and the intensity of shadows or fill light - so the sky and environment merely lights everything the same intensity from all angles, causing diffuse shadows that you’d normally see on an overcast day.

A sliding scale of dynamic range

The highest dynamic range environments are always shot at mid-day with a visible sun, which has not been obscured by any cloud or haze. On the opposite end of the spectrum, a rainy day at dusk is super low dynamic range, easily photographed with a single low dynamic range capture. Of course, the harsh midday sun is not the most popular time of day for photographers or CG artists as it is the highest contrast, and features razor sharp, dark shadows.

The rule of thumb you want to use when looking at HDRIs is that it needs to have a dynamic range or number of stops that matches the conditions. A mid-day sun should be shot with enough brackets + a neutral density filter, so that the resulting HDR actually features a bright sun, sharp shadows and only subtle blue sky lighting. A sunset will require less brackets, and features less range, but is still entirely within the realm of HDRI. An overcast sunset may barely require HDRI at all, other than to capture some extra detail in the clouds near the sun - here we start to venture into the realm of MDRs.

Things to look for in HDRIs in terms of specifications are either "dynamic range", "stops" or “EV” of around 16-18 for mid-day skies, up to 15 stops prior to sunset, and more like 1-3 for overcast, night and dusk situations. Depending on cloud cover, a concealed sun can bring the dynamic range down a great deal. If you’re looking at full environments with landscapes and shadow, you can add more stops to the above guidelines. 

Fixes for MDRs

If you happen to have some HDRIs that are more on the "M" side, you can still account for this and add you own artificial light source, you'll just have to light it as you would a normal scene, using the HDRI for some ambient lighting and colour. This is even applicable if you are using a good HDRI sunset that you'd like to have more punch.

Alternatively, there are two tricks you can use to make any MDR or even LDR (with differing levels of success!) into an HDRI.

1. Copy paste a correctly captured sun into your M or LDR image. Go into Photoshop, select the good HDRI sun with a bit of a feather, copy, load your sky that needs some help, and paste it in. You can adjust the strength and colour of the sun on its own layer to tweak the end result.

2. Make an elliptical selection where the sun is, ensure your image is in 32bit mode, and then boost the exposure (Image>Exposure in Photoshop). Different sizes and strengths of sun and feathering can control the shadow sharpness. As above, watch your colours here as you can end up with a strong coloured light if you aren't careful.

Conclusion

I hope this helped you understand with more clarity what the benefits are to your IBL renders when using real, accurately captured HDRIs, also what to look out for, and how to improve a HDRI that has less dynamic range than you may like.

About

Jay Weston is the principal photographer at Hyperfocal Design, who specialise in HDRI sky environments. His latest project is an ambitious time-lapse HDRI sky dome collection.

You must be logged in to post a comment. Login here.
Thanks, Jay...Question, how did you create the falloff around the spheres, particularly the darker ones, so that they reflect a lot more (almost white-ish) around the edges? I understand a falloff map on the diffuse or something with the fresnel, but I cannot figure how you did this and I like the look of that lighting.
thank you so much Mr.Weston for replying, and this link which u gave very helpful for me in this case, appreciates, I mean atmospheric shadows are when there no direct light only sky is the only light source,
I have a Question for this post....If I need Dusk renders(Late evening ) What kind of HDRI is good.....I don't want want shadows from sun, I just want Atmospheric shadows....any suggestions for this?
For dusk you can use an LDR jpg and get probably about the same range as a HDRI. If there's no sun or moon you might get a couple of stops max, ie no directional or strong light source to cast any shadows. Check out the twilight, dawn, dusk render from this: http://www.cambridgeincolour.com/tutorials/natural-light-photography.htm I'm unsure what "atmospheric" shadows are exactly?
I have a Question for this post....If I need Dusk renders(Late evening ) What kind of HDRI is good.....I don't want want shadows from sun, I just want Atmospheric shadows....any suggestions for this?
I think a lot of the issue is that when we move away from the usual sunny sky images then quite often we can be surprised by what is normal levels of illumination or not. Our eye and it's ability to auto-expose does all sorts of weird things to our perception of right/wrong values and hues. Using our expectation of what the sky should look like as a gauge of the correct values might be wrong too. In practice we need the HDRI to be correct by the original author. If it's not there is little we can do to spot it unless it's really so bad that it's obviously wrong like in the examples above. I bet you could tweak a HDRI with a curves modifier and add perhaps 15% intensity to the mid-range and de-saturate that mid-range a bit and I think no one would ever know it's values were now quite unreal. I'm still unsure about the stars/glare at high f-stops. It might not be noticeable per se, but if you are being so careful as to avoid using directional lights for the sun and instead use the HDR for the sun light source to assure shadows are the correct crispness and the colour balance is perfect between sky ambient light and the sun point light, then the star glare will impact the shadow softness and also cast the sunlight hue into the blue areas of the sky impacting the sky light average colouring. It will certainly be visible in glossy reflections and make pure smooth surfaces appear to have fine scratch patterns to the very discerning viewer. On these very latest cameras with 20mpx+ the f-stops over 8 or so can result in lost details due to something at the photon level (diffraction?!)... so you need to rely on even more ND in these cases for very high MPX cameras/sensors! In the end though, we strive for complete accuracy for the HDRI but we forgive the issues that a star glare may bring about to a loss of accuracy. We may actually want to tweak the lighting any way, because it doesn't meet exactly our artistic needs, so we'll fudge with extra lights or fill or post-processing any way. So despite chasing the perfect HDRI we will probably throw it all to the wind later any way by using some un-real values hehe. Dave
To put it simply I would say that what you are always after with a HDRI is that the dynamic range within the image matches the range in the sky/environment. There's no way to check if this has been done properly until you've loaded the HDRI and inspected it. Asking the photographer for the number of stops or EV in the image is good too. For example if you have a mid day clear sky and only 6 EV, then the HDRI has not been captured properly. I agree that the star/glare isn't the best, but its practically required to get all the dynamic range on bright days. Also I think you'll find that this star/glare doesn't add much if any extra lighting information, and shouldn't effect the sharpness of shadows much either. Furthermore, ~f11 will ensure for most lenses that the entire frame is as sharp as possible. The ideal would indeed be either ND filters made slightly darker, or ISO as you say - even lower than 50, or some crazy shutter speeds!
An interesting article. HDRi are a bit confusing. The fact we can map a range of intensities into any range of bits means either piece of information can be rather useless as a guide for correctness. In the end I suppose all we can hope is that the dynamic range on offer is linearly distributed across the range of bit depth on offer. As long as that is the case then we should get correct results. But how to check/know that is the case seems to be something we can't properly check without already knowing the rough relative intensities available in the shot to start with. Bit of a bummer. All we can ask for is good meta-data I suppose! It's interesting you raise the issue of ND filter for HDR on a very bright day though. As we stop down beyond 11 or so you get a 'nice' star of glare from the sun. Nice as in it looks nice in photos, but isn't good for HDR at all. With the glare 'star' present we're actually saying the sun is a large star shape and not a small circle so it will make the end results inaccurate. So in the end another thing to watch out for with HDRI is the bright spot sun having any kind of aperture diffraction (star/glare) from high f-stop exposures because in essence it's gonna do things like cause softer shadows etc.
gud one ... can u give a realistic render setup... with light and camera setings,,,
Well, this one help me on work so much! Thank you very much!
I like the ones at CG Source. Not a fan of the evermotion ones to be honest.
Might see if Jeff wants to post a link to it in the article as an update.
Just updated :)
He provides them free on his site here: http://gl.ict.usc.edu/Data/skyprobes/ Down the bottom you can download by time of day, by the hour. Might see if Jeff wants to post a link to it in the article as an update.
Great article. Where can one buy the HDRI from Paul Debevec?
Thanks Aldo, your three dome system sounds good - I guess this allows you to easily adjust your sky brightness vs the lighting brightness/power. Sometimes as in real life, the HDRI sky itself can end up too bright compared to the landscape, so this sounds like a good, easy way of controlling that. Regards, Jay
Hi Jay, I tried your HDRI and are fantastic, I love that you do not need any light to support as many artists do, I mean they place a VRaySun or some other source to generate stronger shadows, I currently use a technique that I call TDL(Three domes lights) I use one for light, one for the background and one for reflections and with the help of new options for HDRI in V-ray 2.4 the HDRI work perfectly well without making a false lighting system, congratulations for your work, greetings.
Connor that's a good question. I'm not a colour management expert (I'm still learning myself) but I loaded an 8bit, 16bit and 32bit image into Nuke and looked at the RGB values of a sun. The 8 and 16bit images only go up to a value of 1, while of course the HDRI goes up to 1000's. So I'm guessing it's as you said, there's just more steps in 16bit vs 8bit, but no more range or larger container size. If someone wants to correct me, please do! Richard the VHDRIs are our highest resolution HDRIs. The results you'll achieve will vary based on the exact sky you use. I can look up that file name for the above render if you wanted - maybe shoot me a PM or contact me via my site though. Just wanted to reiterate some wording on the article too - While an HDRI that should by its nature (mid day sun) have a high dynamic range, but does not, this is where the MDR situation is bad for colour casts, etc. You can get sunsets and cloudy days that are naturally more "medium dynamic range" because the sun is weaker or only partially visible behind cloud. Cheers!
Brilliant! I assume I'd want HyperFocal's "VHDRI" series to achieve the results you have?
This is very helpful thanks! General related question, does a 16bit psd have the same dynamic range as an 8bit psd just with more steps between black and paper white?
Hi Martin, Thanks. It can be hard to tell without example renders, and even then, perhaps that example render uses an additional directional light, which can make it harder to tell. If there are some example renders available (note to self, do more of these!), the rule of thumb is probably that scenes with a full, visible sun around mid day to afternoon should feature sharp and contrasted shadows. If they instead have diffuse/blurred shadows as per the example above, then its probably "MDR". Strong colour casts are another clue, and dull highlights. If none of this exists alongside the HDRIs, look for the number of stops/ev etc as I described. If that doesn't exist, I guess just email the photographer/creator and ask them.

About this article

What exactly is the difference between a poor HDRI (or so called MDRI) and a good HDRI?

visibility40 k
favorite_border10
mode_comment21
Report Abuse

About the author

Jeff Mottle

Founder at CGarchitect

placeCalgary, CA