Mantaflow Smoke refinement

Hello.

I’m looking for a way to improve quality of my smoke simulation done in Mantaflow. I’m novice at physics simulation so most of the progress is done by trial and error.

The scene I’m working on is done in real life scale - smoke domain, as the image captured in render is around 18 meters in width. I’ve experimented with Adaptive Domain, unfortunately the cache for it is not working very well. The baking times are long so I want to avoid unnecesarry work and ask here for some advice.

First issue - the smoke looks like its stretched in z-axis. I cannot figure why and all test bakes that I did before had the same problem. Could this be because of the collision object geometry? The smoke is colliding with the ground and trees. The trees are simple cylinders without any subdivision between top and bottom caps. All geometry has all transformation applied:

Second thing is related to smoke resolution. I struggle with the refinement of smoke detail without changing its general shape. I’m looking for an effect like this:


Or this, but without the fire:

Unfortunately after increasing resolution and adding some noise the smoke detail looks very uniform. Here is the render with increased density:

The rest of the parameters:
Smoke Emitter size: 16x2x2m (simple box, no subdivisions)
Smoke Domain size: x-19.5m y-18.9m z-5.8m
Domain settings:


I’m using Blender 2.90

Thanks for any help.

this video may help https://youtu.be/qRqbTZYxRLY
and a shorter explanation(but NOT in english) https://youtu.be/OK_R1l_jY1k

1 Like

Thank you. The first tutorial is very good.

1 Like

Im very interested in seeing your results due smoke/fire being so complex.

Thanks, I’m interested too. To be honest the whole scene where this smoke is located is on the edge of my skills and I’m struggling to put it all together since. Thats why it takes so long, and baking smoke is nearly the last step on the roadmap.
On the bright side Mantaflow got improved during last month, so I hope the game of trial&error won’t be as hard as it was the first time.

Finally had some time to work on this. I’ve rebaked the whole sim with over 500 samples, tweaked noise and collision object parameters. The bake took almost 2 days to finish. It looks better now but it’s far from beeing finished. The biggest culprit is that test baking with low resolution doesn’t allow to make good decision about final sim.
Render with cranked up smoke density and overall brightness:

I’ve done some small scale tests with noise. It looks that increasing Noise Upres Factor yields much better detail distribution.




Closing on the target. The reference for this simulation is piroclastic flow.

The overall shape cannot be affected by effectors to much, or the smoke will fill the whole domain pretty quickly. This can be seen in the post above.

There seems to be hard limit on negative Buoyancy, negative Heat and overall gravity influence. As the effect the smoke cannot behave heavier and flow more near the ground - despite my numerous attempts.

As for the details the solution seems to be high noise level and veeery little vorticity - as this also can stirr the smoke too much.

The flickering is due to denoising - I forgot to disable it before rendering.