Caustics

[jason1025]

Hi Fellow Hashers

 

I believe this was made with AM because about 5 to 7 years ago I saw a guy demonstrating some AM Technics and he had this or something very close to this on his desktop.

 

Can anyone duplicate this? I honestly don't know where to start. This is such a high dollar look yet I see very little examples of AM user doing stuff like this. AM has had the ability to made impressive caustics like this for years yet it seems like the community as a whole has some sort of roadblock.

 

 

Would some like to post the settings to create this or a project, or better yet a Video tut that explores caustics further so that as a community we can start using this powerful tool with better ease.

[robcat2075]

Caustics will be needed to get the colored shadows. For the diffuse illumination a simulated area light will be needed, which means multipass.

 

My model is only a bare approximation of the shape they have there.

 

first attempt, not even finished rendering yet. This is done with a light traveling on a spiral which is not the shape they have used in their image, I suspect theirs is more rectangular, but i had a spiral ready to go. The spiral doesn't create a random enough light placement. more R&D would be needed. But i think something very similar can be had with A:M.

 

[robcat2075]

256 passes finished

 

 

 

yeah , it really needs a rectangular light over it.

[Rodney]

A few years ago J Griffin shared the gems with caustic effects that show up in the A:M demo video.

He was too busy at the time to create a tutorial (still is) but hoped the settings could be useful.

 

See attached project:

JewelsCoverFinal.prj

[jason1025]

Impressive Rob.

 

In some ways your panel looks better than the original. Are you done?

[jason1025]

I suck at this. I cant come up with anything close to that.

 

Can you share your material settings?

[robcat2075]

closer...

 

[zandoriastudios]

Here was something I did in 2003, playing with caustics:

http://www.hash.com/stills/displayimage.php?pos=-697

[robcat2075]

another aborted attempt...

 

[zandoriastudios]

Robert, I think you need a Black background and an index of refraction in your surfaces to get closer to the reference picture

[robcat2075]

It has both. However I would have to make a black sky dome to get black reflections as A:M currently doesn't reflect the camera background color.

 

But first I'm trying to get the shadowing closer.

[NancyGormezano]

It has both. However I would have to make a black sky dome to get black reflections as A:M currently doesn't reflect the camera background color.

 

But first I'm trying to get the shadowing closer.

 

I realize you are trying to use radiosity and get a more accurate result - but I have to wonder if you've considered fiddling with light lists, 4 kliegs with colored zbuf shadows (Diffuse=OFF? spec=ON) to augment the look? I don't know if it would help.

[robcat2075]

closer...

 

 

A more random path for the area light would solve the stepping in the shadows.

 

 

Part of the challenge of duplicating the example is that I don't know much about how the lights are arranged, I can only guess from the shadows and reflections.

 

I'm not a lighting guru, but I think the essential elements of that imagery is available in A:M. More study of the original image and tweaking of the A:M version will certainly get closer to the goal.

 

for further investigation...

 

 

-their shape is much thicker than mine and has a very specific beveled shape to the edge with probably some random non-uniformity to break things up

-I've only approximated the colors of their model.l

-I'm not sure how they maintain the sharper caustic patterns with the hazy illumination. That would need some more R&D

-My version is an all-at-once attempt. With compositing of different lighting situations many of the more nuanced details in the original could be had.

 

 

Here's the PRJ to play with. Click thru any image requests.

 

CurvedGlass08__Rays.prj

 

"Rectangle" is the path that "Spiral Dome Light" travels to simulate an area light. With multipass ON the light travels the path from 0 to 100% in the first 0.2 of the frame

 

 

I realize you are trying to use radiosity and get a more accurate result - but I have to wonder if you've considered fiddling with light lists, 4 kliegs with colored zbuf shadows (Diffuse=OFF? spec=ON) to augment the look? I don't know if it would help.

 

that would also be interesting to pursue.

[robcat2075]

Thicker slabs looking promising...

 

(see next post)

[robcat2075]

Bevels need to be fatter and the sides need some sort of ripple to make more interesting refraction artifacts.

 

[robcat2075]

The challenge of this simulated area light thing is making a path for the light to traverse that somehow evenly distributes the light within the shape. of course that's not truly possible with a "path".

 

The more passes you have in your multipass render the more accurately the light is describing the line and not randomness. The more instances there are of the light moving in a straight line, the more obvious the stepping is in the shadows.

 

 

Any math whizzes know how to truly randomize points within an arbitrary shape?

[robcat2075]

This one is too bright, but you can see telltale signs of the zigzag pattern the light is on in the "teeth" at the edge of the blue shadow.

 

[robcat2075]

Looking at the original image some more, part of that "look" is that there's a bunch of detailed crap off-camera that's being reflected in the plastic pieces. It's not just the refraction artifacts that we are seeing.

 

We'd have to know what that is better to get all that little detail.

[NancyGormezano]

Perhaps try an environment map using some interesting image...maybe even your rendered image? I like how it's coming along.

 

I think it interesting, but crazy to try to copy exactly something that I'm betting was most likely done with cheats, tricks, in some other software. Their algorythm is also an approximation. Not necessarily a more accurate method. Even when commercial images are created using photography, there are always cheats (need I state the obvious: photoshop photoshop photoshop) to get a certain snazzy look, that is not necessarily reality, nor what the camera saw, but more what they wanted the camera to see.

[jason1025]

Hi Rob

 

Thanks for this. I think you did a great job. Can you post your most recent version?

[itsjustme]

Great stuff, Robert!

[robcat2075]

I hereby declare this... pretty close.

 

Aside from the off-screen reflection items mentioned above ( I think that's a lightsaber showing in the yellow block), the shadows are still a bit too bright and the blocks need a smoother transition from the faces to the bevels. Those are doable but I'll let others pursue that. However, I think this shows we can do those sort of renders with A:M.

 

I added some bump to the surface; that can be turned off easily in the material.

 

256 passes with a denser, less pointy light path, 5.5 hours:

 

CurvedGlass11h_denser_path.prj

 

render preset: CurvedGlassTest.pre

 

 

 

Reminder: when doing caustics scenes do a Q render before trying a Shift-Q (progressive) render. Then you can often try some minor parameter tweaking while a progressive render is running.

 

If you make any major changes to your scene, save it then reload it to completely reset A:M's awareness of what's in it.

 

 

 

I think it interesting, but crazy to try to copy exactly something that I'm betting was most likely done with cheats, tricks, in some other software.

Yes I think the original image was carefully stage managed and manufactured and not the result of just a few minutes of work.

[NancyGormezano]

looks great.

 

I took your original project (not the one you just posted) and added Matcap shader (diffuse, and ambiant with ambiant intensity = 200%) to your blue glass material (first image - only 3 pass), then added a dome with image projected on it to give the plastic something to reflect (2nd image - also 3 pass). Also changed the reflective blend to 50%, made ambiance color black. Made ground not reflective

 

Different images were used for the matcap, and the dome.

[zandoriastudios]

I don't know if the original was made in A:M or not...It could just be some default colored glass in a different program. I'm at work, so I just made a quick version in Rhino and rendered it in Vray:

[Gerry]

This is all dazzling! Great testing everyone!

[jason1025]

Why is it that Rhino and the original look like glass and the ones we did in AM look more like plastic? Is it because of the swirly effect of color along the edges?

[robcat2075]

I think the edges being darker than the faces has something to do with it. that's something I'd try to pursue if I were going to pursue it.

[robcat2075]

The edge detail is important too. if I turn up the refraction that adds some.

 

I'm not a lighting person. Do not regard me as the last word on what is possible.

[NancyGormezano]

I think the higher the spec, the higher the reflectivity, the more it looks like glass?

 

(all done with 3 pass, so don't look at shadows)

[HomeSlice]

Is there a step-by-step tut for radiosity somewhere?

I've been reading through forums posts for an hour and I still haven't found one. Most posts on radiosity assume you know some things that I obviously don't.

I keep getting bright splotches in my renders even though I've tried upping the photon to 100K and the photon samples to 5000. I've also been playing with the sample size, but I can't get rid of the splotches

 

[EDIT] OK I found a great introduction here:

http://www.hash.com/forums/index.php?s=&am...st&p=327730

The only thing he didn't mention is to right-click on the chor and choose Calculate Radiosity.

[robcat2075]

btw, I noticed that ray traced lights and z-buffered lights give somewhat different results with caustics. The very curious will want to experiment.

[dblhelix]

(HomeSlice,

search +light +tutorial by ypoissant.

cornell box had the project in there, i was only reading so don't know if the link worked.

it's not step by step, lack of images is a problem but the text is helpful.)

[John Bigboote]

I've been watching this thread... hoping that it would bear some fruit. My first reaction was 'no, A:M did not make that image.' I have recently tried to recreate an image I have as a desktop image(BELOW) into animation using A:M and have been similarly frustered by caustics and refractivitity, as well as render quality. That one image Will posted looked really good... and I am intrigued by Rob's 'raytraced lights' idea... maybe will try that, along with radiance- which I didn't try mainly because I was going after animation and it might be too render intensive. ANYWAYS... thought I would share my attempt:

 

THE IMAGE BELOW WAS NOT MADE BY A:M... THE MOVIE WAS.

glass_caustics.mov

[robcat2075]

THE IMAGE BELOW WAS NOT MADE BY A:M... THE MOVIE WAS.

 

Hay, matt... part of the glass look in the image is a using a gradient for the reflection value. Surfaces seen more flat to the camera will have less reflectance and surfaces seen on edge will have more. Try that.

[robcat2075]

THE IMAGE BELOW WAS NOT MADE BY A:M... THE MOVIE WAS.

 

also... use the "Glossy" specular plugin shader for the little spec of highlight that is on the lower left edge of the lozenges.

 

I made a brief test and the initial problem i had was getting the reflection of the underside of a lozenge to be dark rather than show a reflection of the bright topside of other lozenges.

 

sudden thought... there might be some application for that new reflection shader in solving that.

[jason1025]

THE IMAGE BELOW WAS NOT MADE BY A:M... THE MOVIE WAS.

 

Hay, matt... part of the glass look in the image is a using a gradient for the reflection value. Surfaces seen more flat to the camera will have less reflectance and surfaces seen on edge will have more. Try that.

 

 

are you talking about this?

FresnelReflection.mat

[robcat2075]

wait I think i got the reflection problem solved.

 

How about this, Matt? This is closer to the appearance of the disks in the sample image...

 

[robcat2075]

the reflections could still stand to be darker but I think this is very much in the neighborhood...

 

[jason1025]

the reflections could still stand to be darker but I think this is very much in the neighborhood...

 

 

 

Thats looking really good. The original almost looks like there is some cloud cover. Vs your shot which looks like the sun peaked through.

[jason1025]

Could I study that project Robert?

[robcat2075]

I made my disks by squashing a sphere, but I think mine are too skinny.

 

 

lozenges0C.prj

[John Bigboote]

Wow, great progress Rob! I have been involved in an 'non-computer' animation job that has me doing some old timey pixillation so I have not been looking-in on the forum as often as I like to... so sorry for late response. I will need to digest your project to see just what the hey you got going on there... but it is sure looking better than mine, and I put a lot of time in on that chasing my tail, so I appreciate your effort! I see you are getting a better edge-render too, that had me concerned...

[robcat2075]

I have been involved in an 'non-computer' animation job that has me doing some old timey pixillation

 

We definitely want to see that when it's done!

[robcat2075]

Thats looking really good. The original almost looks like there is some cloud cover. Vs your shot which looks like the sun peaked through.

 

I notice that in the original the white background is not stark-raving white like mine but a more nuanced gray gradient. That's probably important.