Forming an Object (or an image) with fluids for

[lord-mcgoat]

I am working on a commercial that requires transformation of smoke into a particular object (or an image). I tried to reverse the cache but it does not look good, due to the fact that swirls give it away. I would love some input on how to approach this problem. Any ideas? Duncan any suggestions?

[duncanbrinsmead]

Perhaps you could use the object as a fluid emitter then emit smoke, run a while and set the start state to a position where smoke is in much of the grid, perhaps animating the emission to zero, so the original emitter is then not too noticable. Then rather than cache and reverse, run the simulation from this state and gradually increase the dissipation while at the same time increasing the emission from the object up from zero. The smoke will disappear from the grid except where it is nearer and nearer the object.

You might have some success by running part of a backward cached simulation and crossfading it into a forward run simulation. Rather than composite you could overlap two separate fluids.

There is a different approach, but one that is not currently supported in Maya. Jos Stam, who wrote the Maya fluid solver, co-authored a paper on goal oriented dynamics( sort of inverse kinematics for dynamics ). One could take several key objects and have the fluid go through plausible dynamics transitioning from one key state to the next. The original paper had a method that was extremely slow to compute( requiring iteratively simulating the entire simulation for each frame ). He later did a paper that had a much faster, and more potentially practical method. Hopefully we can add it at some point, although the UI and such would need a fair bit of work and thought.

Duncan

[lord-mcgoat]

Duncan

Thank you so much for your advice.

The solution that seem to work for me is a combination of straight animation from a fluid emitter comined with the method you presented at Siggraph last year, where you used the colorAtPoint and add it into voxels.

I design my own version of image sequences that gives it
a "fluid" initial push and then I let dynamics to complement
it.

I found it to be a great combination of artistry and technology.

I am still learning fluids, and exploring new ways to use.

The aspect that I find it difficult to control is the detail in the smoke - the scale of the grain. I am trying to match it to the footage and have relative success.

If you have any more advice i would love to hear it.

P.S. I have read the papers you have mentioned a few months ago, but since I am not a C++ programmer, I could not think of a way to utilize the data in MEL or expression.