Cut Crystal Tests

[WytRaven]

I have decided to make this a proper WIP thread and will continue to post progress images as I attempt to perfect cut crystal glass.

This scene is a test of several things at once:

• Worn marble texture under sun lighting
• SSS liquid (coke in this case)
• Modelling realistic cut crystal glassware
• Brushed metal surface

This render was unintentionally halted at 17.5 hours and upon resume the micro seconds / sample had dropped from ~18 to ~3; I figured something was wrong and so stopped it. It's all good as I had achieved what I wanted from the render anyway.

I originally got into 3D rendering because for as long as I have memory I have always been facinated by the reflection and refraction of light. As I teach myself the fine art of 3D modelling I am always leaning toward objects that either bend or bounce light, preferably both. Cut crystal has got to be just about the epidimy of that.

I use Blender as my primary modelling tool and generally model using polys only, no nurbs, no subdiv, just polys. Recently I started playing around with MoI and was stunned by the ease with which one can create complex shapes. I decided it was time to tackle my holy grail of cut crystal and proceeded to get down to modelling a tumbler in MoI. Within about 30 minutes I had built WytRaven's Cut Crystal Tumbler Mk I. I exported into .obj (after screwing around for about another 30 minutes with the tesselation parameters). Once in blender I was appalled to find my lovely tumbler was a whopping 150,000+ polygons; dirty,messy polygons at that. I did a test render and wasn't overly impressed with the results.

I went back into MoI and built a second tumbler, much better design this time but still exported to a bloated and messy poly object. Now don't get me wrong here MoI does a pretty amazing tesselation job under normal circumstances but once you start performing lots of boolean operations etc it quickly gets confused.

Frustrated by these inefficient results I decided to bite the bullet and do what I do best and build my tumbler from good old hand wound polygons. 8 hours of painstaking modelling later I am very happy with the results. I love polys and these tumblers stand testament to the validity of my love. Not a single superfluous polygon to be found here

These tumblers were built without the use of any booleans or any other such 'tools' just hand placed vertices. Ok, yes, I am a control freak and I'm proud of it!

Original render 1920x1200 (50% reduction shown) and last time I saw the stats it had reached around 1600 samples per pixel. I was going to let this go much longer...

Who needs 150,000+ polys ???

[Kram1032]

very nice render
very long description O.o
very perfect brushed metal !
very interesting caustics

[WytRaven]

very long description O.o

I can be a little long winded sometimes... ok all the time

Hey I figure explaining a thing or two might help others who are trying to learn...well that's my excuse and I'm sticking to it

[Kram1032]

I'd help you
What do you want to know?
(except, if it's MoI related - I don't know that at all)

[Whaat]

Hey WytRaven,

Your glassware is incredible!!
Care to share the models??

One thing to be careful of when you are rendering liquids is to check your gamma value for the absorption and scattering coefficients. If you are using the values from the WIP Indigo manual, make sure gamma is set to 1.0. Cindigo might be exporting gamma of 2.2 for these values. I am not sure.

Great render!

[WytRaven]

Thanks for that gamma info Whaat. I had just started a new render so I'm glad you told me that before it was 30 hours in

I was concerned that the coke was looking a bit more like zambuca than it should in the first render, now I know why

I would be more than happy to share the models once I'm completely happy that they are perfect but I would like to complete the set first. Next up some cut crystal plates

[Kram1032]

cool
So, is it 2.2x denser physically correct coke?

What are your overall settings for bounce depth, MNCR and so on? - You might get some more improvements, by tweaking them^^

[WytRaven]

From memory (i'm at work) depth was 5000, mncr 750, BDIR MLT.

I've set up a new render without the balls, fixed coke, and zoomed much closer in on the tumblers so I can really see the detail properly. New render is using a little of Cool's testing work and is running with a depth of 10000, mncr of 30000, BDIR MLT.

BTW does anyone know a correct cauchy b value for Crystal? Would love to try these with chromatic abberation

[BbB]

Hi Wytraven,
Great stuff there. Yeah, there's a lot of trial and error involved in converting 3DMs to OBJs. Is it possible to render Indigo images straight from within Rhino, for instance? I'm told Maxwell has a good plugin.

[WytRaven]

Ok so I have fixed a few things, upgraded to indigo test 8, and have managed to calculate a reasonably accurate cauchy b coeficient for crystal. I will post render progress images as I go starting with a render at 2 minutes in.

Here are the key render setup parameters to begin:

Code: Select all

<renderer_settings>
  <width>1920</width>
  <height>1200</height>
  <metropolis>true</metropolis>
  <max_num_consec_rejections>1500</max_num_consec_rejections>
  <ray_origin_nudge_distance>0.00001</ray_origin_nudge_distance>
  <bidirectional>true</bidirectional>
  <max_depth>15000</max_depth>
  <display_period>20.0</display_period>
  <image_save_period>120</image_save_period>
  <super_sample_factor>2</super_sample_factor>
  <hybrid>false</hybrid>
  <cache_trees>1</cache_trees>
  <num_threads>2</num_threads>
</renderer_settings>

Render @ 2 minutes (JPEG compressed) 50% scale: Click for full res PNG

[WytRaven]

Render @ 10 minutes (JPEG compressed) 50% scale: Click for full res PNG

[WytRaven]

Render @ 30 minutes (JPEG compressed) 50% scale: Click for full res PNG

[WytRaven]

Render @ 1 hour 30 minutes (JPEG compressed) 50% scale: Click for full res PNG

[WytRaven]

Render @ 4 hours (JPEG compressed) 50% scale: Click for full res PNG

[WytRaven]

Render @ 8 hours (JPEG compressed) 50% scale: Click for full res PNG