View Full Version : Controllable, realistic fire and smoke in RT

09-13-2006, 04:42 PM
It's been a long time since I was at CGS, but I hope there will be some support. It concerns the following topic:

For an extinction training simulation I have to find ways to virtually burn objects in a video sequence. The idea is to create virtual fire and smoke and "overlay" it with the video. Edge detection will be used later to define areas where fire is and where not.

Video is controlled by LabVIEW (National Instruments). Fire and Smoke has therefore to be totally controllable by LabVIEW too. For this reason the best deal is to have fire and smoke that is formed or built parametrically (speed, orientation, dead, size, ...).

Primary goal is to have burning that looks at least realistic as real burning. And it has to be in realtime!

My task atm is to find different solutions and technical approaches and compare them. I've read a lot about computer generated fire and smoke and these are the keywords that came up:

Navier Stoke equations
flame shaders
Lagrange vs. Euler
Procedural shaders
Volumetric Fire
grid based (solvers?)
particle systems
also point sprites (they used it in Lord of the rings I guess, it looks awesome)
DirectX, High Level Shader Languange, ATI's Adv. Shader Language Interface, Renderman
Well, these are just terms that spook in my head, so they are not sorted. I gonna build three groups of different attempts:

Physical based solutions with its solvers. Fluid dynamics
Procedural based solutions using noise or fractals
3D or 2D Particle systems
I tend to use DirectX for implementation. I know that most shaders are written in Renderman, but they can converted into other languages. I'm not sure if physical based solvers work in realtime, but results look awesome (stills from Fedkiw or Musgrave, Ebert, etc.).

Well, to cut a long story short, do you have any ideas that are quite simple (calculation or programming) and are fast (realtime application). Don't forget the total controlability.

Thanks guys for helping me.

09-14-2006, 01:11 PM
check out some of the papers on this site:


they may not be real time but they're interesting. Also look at the websites of the R&D developers for the big studios like pixar - likelihood is they've done something similar in the past and published a paper on it

09-15-2006, 09:44 AM
These are some good papers on fire:

http://www.cs.brown.edu/~tor/sig2002/StructuralFlames02.pdf (http://www.cs.brown.edu/%7Etor/sig2002/StructuralFlames02.pdf)


http://graphics.ucsd.edu/~henrik/papers/fire/ (http://graphics.ucsd.edu/%7Ehenrik/papers/fire/)


PS it would be great to hear which approach you end up using

09-15-2006, 01:10 PM
incidentally, has anyone tried any of these methods (or similar) using a Maya Fluid grid as the navier stokes solver?

I was looking at Level Set/Implicit Surface methods that combine an NS grid and wondered if I could decrease dev work by just using Maya's own grid. Then I could retrieve values from it to use for my particles and level set. Rather than starting from scratch and wasting lots of time building my own NS solver.

Any tips welcom :-)

09-15-2006, 08:50 PM
What about using Lattice Boltzmann for solving fire as it seems to solve gas and fluids well? Not sure how much info there is out there though ...

09-20-2006, 12:00 PM
Thanks so far for your comments. I was searching a lot and found many papers, but those from rendermaniac are new for me :)

Concerning Level Set Methods and Implicit Surfaces, I have got the book from Fedkiw.

The thingy is, I'm not going to use Maya or similar, thus I have absolutely no plan how to use solvers (lattice, grid, you can name it) otherwise ...

CGTalk Moderation
09-20-2006, 12:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.