rendering fire, test renders

[nicoladanese]

Hi all!

I'm doing some research on rendering pyro explosions, and here's the result of my comparison between mantra on houdini and renderman 21 on maya using a vdb cache from a pyro explosion.

https://vimeo.com/218752467

my thoughts

I'm pretty happy with both of them, fire looks detailed as I expected from the hi res sim, and I like the way the light scatters through the smoke and lit it. renderman version came out "hotter" in terms of color than the mantra one, but I guess is something to do with the way blackbody radiations infos were passed to the emission component of the shader, so I can probably get closer one to the other by remapping values in a slightly different way

mantra

pros render the explosion with mantra was fast in terms of setting up the scene, and the shader is much easier to play with. It's also nicely setup when you create the pyro sim, so you can get a nice looking fire in a few click.

cons I found render time slower than renderman, and I couldn't find a way of getting rid of the huge amount of noise in the meshes lit by the explosions (especially when the core of the explosion is brighter) , although my render settings are pretty high (I think): pixel samples 9/9, min/max ray samples 1/9 with noise level 0.005, diffuse and volume limit 5

renderman

pros I managed to get faster frames with less noise on the surrounding environments

cons set up the shader was kind of a pain compared to mantra. I needed to import fields and remap some values to make them work with PxrBlackBody, nothing too complicated but far more slow compared to mantra

 

I'd really like to get feedback, ideas, personal experience or anything you'd like to say about it  

[schwungsau]

for mantra you should try increase the minimum samples... ray samples settings like 3/5 may give you less noise...  also diffuse quality have a big role in reducing noise with volumes.

[nicoladanese]

hey Heribert! thanks for your reply

I tried min/max samplese to 3/9 and diffuse quality 10, definitely got rid of 90% of the noise, but the render time went from 25 minutes to more than 3 hours and a half! too bad you end up spending most of render time rendering the environment rather than the fire itself...I wonder if there's a smart way of converting the fire into something a little simpler to lit the env without using the actual fire cache...some kind of emissime mesh tinted with the color of the fire...

[derjcmp]

Increasing the stochastic samples to something like 12 reduces alot of my render noise. You should give that a try aswell. Also for my volume quality I set it to about 0.1 and my shadow quality ti 0.5 

[schwungsau]

7 hours ago, nicoladanese said:

hey Heribert! thanks for your reply

I tried min/max samplese to 3/9 and diffuse quality 10, definitely got rid of 90% of the noise, but the render time went from 25 minutes to more than 3 hours and a half! too bad you end up spending most of render time rendering the environment rather than the fire itself...I wonder if there's a smart way of converting the fire into something a little simpler to lit the env without using the actual fire cache...some kind of emissime mesh tinted with the color of the fire...

oh you using the volumes to light env. !   this is always expensive to render.....

a common solution is, to mesh out the heat or temperature field as simple poly object give it a orange/yellow texture and use it to light the envr.  looks pretty nice and render a hell lot faster.

btw. the only "rea"l solution is this here:
http://www.cemyuksel.com/research/lgh/lgh-high.pdf

 

 

 

Edited May 30, 2017 by schwungsau