PDA

View Full Version : Lighting, Caustics, Shadows...



IMI
12-22-2007, 05:00 PM
I've been doing some tests of caustics, and getting such wild variation in results, I can't tell what looks right anymore. I've been at this most of the day. What started as a picture I wanted to make involving a champagne glass has become an obsessive exercise in frustration. I figured I'd start with the glass and after I got that right, add more to it. Ignore the wood texture - it's just a low-res crappy jpeg I threw in there so I could see it against something more interesting than a solid color. I threw in the sphere to have another transparent object to gauge the caustic results against.

But, the lighting is changing things so much. The first pic on the left is using a fairly large area light. caustics seem to come through OK, but I'm not so sure the shadow looks right.
The middle upload has the light changed from Area to Distant. The shadow looks a little better, but still doesn't seem right. And although there isn't much of a difference in brightness, there's a definite difference in caustics. The pic on the right is an area light again, but sized way down. Again, shadows look wrong, and the caustic effect is diminished.

The objects all have a mixture of the Dielectric and Standard nodes. The glasses have a Crumple node added to the bump from the alpha, with a very small setting for the amplitude. The sphere has a higher refraction index, and no bump.

I haven't changed the caustic settings or anything else in any of these renders, just the light type and size. The light positions have stayed the same.

Is there some special trick to getting glass objects to render with more realistic caustics and shadows using the lights LW has? I place a glass on a table with a bright light above it, and get nothing like this. The shadows are never dark, and fade off, instead of staying even like these pictures show. The caustics I see on my real-life table appear to be doing what caustics are supposed to be doing.

It's aggravating. With Dielectric we can finally get really nice glass, but that doesn't do much good when the lighting doesn't seem to be cooperating. I tried it with radiosity too, which seemed to erase all the caustics and shadows altogether.

I hope it's just that I'm doing something wrong, not setting my lights up properly, or maybe missing some way of controlling a shadow falloff or something.

LW 9.3, 64-bit
LightWave renders, not FPrime or anything else.

Any suggestions would be appreciated. :)

IMI
12-22-2007, 05:03 PM
Also, yes, I have raytrace shadows, reflections, transparency and refraction turned on in the Render Globals panel. I forgot to add that and didn't know if editing the post would mean I'd have to re-upload the images.

Maxx
12-23-2007, 06:07 AM
LW Caustics need a fairly huge overhaul, so the differences and inconsistencies that you're seeing really aren't that unusual, as I understand it. (I tend to leave caustics off and just hope no-one notices. Actually, now that I think about it, I wonder if my old version of HD-Caustics will still work? hmm.. :cool: ) I have, however, read - I think on the board here, though I can't point you to a specific thread right now - that using luminous geometry in conjunction with GI on Dielectric surfaces can actually trace true caustics, but it may take a bit of time. However, if you're obsessed, it might be worth checking into.

serge
12-23-2007, 06:25 AM
Caustics don't work with distant lights.

IMI
12-23-2007, 06:36 AM
Thanks for your input, Maxx. I was sort of hoping I was doing something wrong, not that LW can't handle it. ;)
I can get good caustics all day long using spheres. They seem to lend themselves to it. So if I ever need a crystal ball in a scene, I'm in there. ;)
Just can't get it to do well, consistently with any other kind of object. Maybe a box - haven't tried it yet...

I also didn't mention above that I wasn't getting ank kind of good results at first. Then I sized everything way down, and it started looking better. The materials and lighting didn't change much but it seemed to make a big difference with the caustics.
I'll try a search and see if I can find anything out about the luminous+GI suggestion. I did finally manage to get a halfway decent look to my glass shadows, good enough for my purposes at least, by using more strategically-placed lights with different falloffs to "erase" the shadow and make it a little more realistic.

HD-Caustics? never heard of it. If it's old though, it wouldn't work in 64 bit LW.

Andy Webb
12-23-2007, 06:37 AM
Shadows never look right when Caustics are turned on.

For some reason shadows just become flat areas and loose any subtlety.

I have to say Caustics really need a rethink in LW.

IMI
12-23-2007, 06:37 AM
Caustics don't work with distant lights.

It would appear not. :(

IMI
12-23-2007, 06:40 AM
Shadows never look right when Caustics are turned on.

For some reason shadows just become flat areas and loose any subtlety.

I have to say Caustics really need a rethink in LW.

Yeah, it's mighty frustrating when you *think* it should be working, but it's not, and you don't know if it's you or the program.
I'd definitely agree that caustics needs an overhaul.

Maxx
12-23-2007, 09:42 AM
I've also read elsewhere that people have had decent success by placing an extremely bright point light set only to effect caustics (no shadow, diffuse, or OGL) near the object intended to cast the caustics. It's something else to look into until NT re-writes the code, which is hopefully soon.

IMI
12-23-2007, 01:11 PM
Well, I tried your suggestion regarding the bright point light. The results are...interesting.

I quickly tossed together a couple of fairly thick glasses and lit them from the front left with a very large area light casting shadows, from the back with a medium, no-shadow spotlight, and from the same spot as the area light with an extremely bright point light, affecting only caustics.

It may not be accurate, but it sure is... pretty. :D

I also used a background image in Image World as a light probe, and Monte Carlo radiosity to diminish the effect of the area light shadow. The shadow still doesn't look right, but I'm rapidly coming to expect that from LW when dealing with glass.

Well, if nothing else, I've learned a whole lot about the Dielectric and Standard nodes in the past couple days. :D

toby
12-26-2007, 01:46 AM
your area light shadows actually look more accurate than the distant light shadows, remember that if the light that's going through the glass brightens up the shadow evenly, then that means it's not getting distorted into caustic artifacts, so with accurate caustics you should see a partially opaque shadow.

And remember this is exactly what reference images are great for!

As for LW being accurate, I just did a caustics test with a glass, one light at 100%, and found a pixel with a value of 10,916,385.16%. Yes, 8 digits - over ten million. Highest value I've ever seen in an image. I'd say it needs some work :)

Nangleator
12-26-2007, 07:47 AM
... a pixel with a value of 10,916,385.16%.
What would that do to the next bounce calculation in radiosity? :hey:

toby
12-26-2007, 09:47 PM
Think 'ground zero'
:cool:

IMI
12-27-2007, 03:43 AM
As for LW being accurate, I just did a caustics test with a glass, one light at 100%, and found a pixel with a value of 10,916,385.16%. Yes, 8 digits - over ten million. Highest value I've ever seen in an image. I'd say it needs some work :)

I'd call NASA next time that happens. Sounds like you've created your own little supernova. ;)

I've discovered the Occlusion node works well for making the shadows a little better. Takes a little longer to render, but does a decent job of faking it.