Lets talk about Mantaflow


(Gilles Charbonneau) #563

Good suggestion, it helped, but now the fluid is going through the effector, I tried to put a higher distance number, same thing, I think that Mantaflow is great for volumetrics, works like a charm, but it is still very buggy with fluids as far as I can tell, but I will keep an eye on it, this is very interesting development for Blender!


#564

Mantaflow branch recieved new update: https://developer.blender.org/rBfc31bd2950e3a0efe51da33bdeaba60653171f13


(Gilles Charbonneau) #565

Cool, and thanks for the info, now we need someone to compile it, my last attempt wasn’t much of a success!


(fin.eskimo) #566

Here’s a WIN64 build with the latest commits


(fin.eskimo) #567

also… here’s a short sim I put together… just thought I’d drop link…


(Gilles Charbonneau) #568

@fin.eskimo, thanks for the build, works great, many glitches fixed, only problem so far is that it crashes when I attach an icosphere to the particles and I try to save, Blender just exit, anyone else with that problem?

BTW, and if you haven’t figured it out yet, you can export the whole thing to Alembic by simply adding an empty modifier to the domain, like a smooth modifier with zero values, it will export the mesh and the particles as points, you then use duplivert to assign a mesh, like an icosphere, to the particles points, works like a charm!


(fin.eskimo) #569

Yup… I can reproduce that crash… if I find a workaround I’ll post it here…


(Gilles Charbonneau) #570

My workaround so far is to export to Alembic and assign the icosphere using duplivert, its actually pretty straightforward, but my guess is that this is not how it is intended to work! :slight_smile:

Yep, the penetration problem has been fixed, loving this, great work! :slight_smile:


(Gilles Charbonneau) #571

Yet another Mantaflow beach test, this time at a resolution of 256, which took about 70 minutes to simulate on an I7 2700K, I think that anything under 512 lacks realism, bit still, it does look good!

https://vimeo.com/279171145


(zeauro) #572

Problems that I reported have been corrected.
We can bake smoke or liquid at higher resolution.

My only problem is that I did not succeed to make fluid guiding work.
I ended up with non moving smoke or crashes (if fluid guiding is disabled and there is a guide object in scene, attempt to bake smoke domain will crash the build).

@sebbas , can you detail expected workflow.
Which data has to be baked, first ? What kind of fluid guide animation is supported ?


(Gilles Charbonneau) #573

Simulating a beach test at a resolution of 512 at the moment, and it seems to be working just fine, brilliant, might take a while though, but I will post a render when done!


(oaschwab) #574

I wanted to test times and results with an animation I made with elbeem but I could not get anything to show up. It’s a very large scene and when I opened it up it had already transferred all the settings to mantaflow. I tried baking and it finished in 0.2 seconds. I tried removing all the mantaflow effects and reapplying them. It would take time to bake but nothing would show up in the render or preview.


It is a very large domain, is there any limitations or problems with large scenes like this?

(Gilles Charbonneau) #575

Have you set the clipping of the camera accordingly, it may seems stupid, but often that is the problem in large scenes?


(oaschwab) #576

Ok I completely redid the scene from scratch and it works. Kinda, maybe. I don’t get a preview. But when I render it appears to have smoke and fire applied to it but the volumetric shading is intense. Are the attribute names the same for mantaflow as they are for elbeem? density and temperature for principled volume?


(sebbas) #577

Yes, fluid guiding workflow is slightly different now.

To use the guiding, one first has to define the guiding velocities. These can come from one of the following:

  1. an already existing (baked) fluid domain. In this case the (main) velocities from that domain are just referenced. In the UI this can be achieved by going to the guiding tab, setting the source to “Domain” and selecting a fluid domain (must be baked already) as the “guiding parent”.

  2. one or more guiding objects (in the UI they can be found next to the collision objects). These objects should be animated and pass through the domain (otherwise no velocities would be captured). Once the animation/ movement of those objects is fix, the guiding has to be baked (so that guiding velocities from our objects get written to disk).

Once the guiding velocities are set up through one of those ways, the actual simulation baking can begin. This works in the usual fashion (-> “Bake Data”, etc.).

However, the 3 parameters from the right column of the guiding tab (weight, size, factor) can now also be used to customize the strength / look of the guiding. That is, during any data bake they will be taken into consideration. For smoke those parameters currently make a little more sense, for liquids they will work too but I also have to play around a bit more to see how and if they make sense “artistically”.

And one last thing: Note that the main bake is much slower with guiding enabled. This is expected behaviour and just something to keep in mind.


(sebbas) #578

@zeauro And yes, thanks for pointing out that bug. Just pushed a fix with some more sanity check to prevent crashes.


(zeauro) #579

I am trying to use a guide and it does not work.
Bake Data ends up with a non moving smoke.

Console indicates an error :
Traceback (most recent call last):
File “”, line 1, in
File “”, line 103, in bake_fluid_data_122
File “”, line 96, in bake_fluid_process_data_122
File “”, line 46, in smoke_adaptive_step_122
File “”, line 102, in smoke_pre_step_low_122
NameError: name ‘guidevel_s122’ is not defined

I have guidevel_framenumber.uni files into guiding cache folder. But it looks like blender have difficulties to read them or they are invalid.


(Gilles Charbonneau) #580

A beach test, resolution of 256, needs at least 512 IMHO, no particles, as I cant save them for now, also, I am trying to get rid of the voxel steps in the mesh, but all in all not too shaby!


(Bernardo) #581

the layering problem where there’s the thin layer of water retreating it’s still quite visible. Same problem happens also with the FLIP fluid sim, I’ve been told it’s quite common with flip particles simulations. Correct me if I am wrong, but I don’t think higher res is gonna fix it. Anyone knows if devs will be able to fix this issue in the sim?


(Gilles Charbonneau) #582

I think higher resolution would only make it a bit less noticeable, the effect it a lot less visible, if at all, on flat surface!