Anyone using Lucid in production?

Posted by:vklein
Data created:9 June 2016

I'm asking, because of the 3 projects I wanted to use Lucid on, in no one Lucid really delivered a somehow usable simulation.

Cloth gets stuck in geometry, fluids rinse particle by particle out of containers. Way too many values are dependent of each other, for instance changing "Sub Steps"/"Iterations" in order to minimize collision errors completelly change the viscosity of fluids and cloths. Some "Particle Radius" works better for fluids, but the cloths in the same simulation then makes errors, and vice versa. We lost valuable production time fiddeling with Lucid errors, completelly negating it's simulation speed, being in the end a lot slower than using other slower but more stable solutions (Maya Cloth/Bifrost). We regret the 2 licenses we brought, In my opinion Lucid is not usable on real world productions, only on entertaining 3d artists between the real work by toying with really fast but highly error-prone physics.

The parameter inter-dependency and constant radius size is a limitation which in a way allows the speed to happen. We have to work with NVidia Flex simulation, which is what Lucid uses, and these are some of its limitations. That being said, however, cloth should not be stuck inside geometry and I would like to look at an example of that.

Marsel Khadiyev (Software Developer, EPHERE Inc.)

Yes I will, ranting without examples is of cause bad style. I have to clean my scenes from confidential geometry, my clients are a little paranoid.

Im going to be trying it out on a commercial animation for softbody stuff in the next few weeks. Ill report back with how I go. 

So here is a test scene of an industrial washing mashine with water interacting with cloths. Simply simulate it won't take long. 

The water looks nice (would increase the particle count for final resolution by 2-4x), but drips out of the container. The fabric interpenetrate badly and at about frame 360 gets stuck in the geometry.

We tried to simplify the geometry, build it from multiple convex boxes and other tricks, but in the end the dripping and cloth being stucked in geometry remains, so we switched back to the original mesh.

Because of the dependency of "Sub-steps", Resolution and Viscosity you simply can't turn "Sub-Steps"/Iterations to high to try to get rid of the imperfect collisions, because then the water becomes thicker, like sauce and the cloth behave like thick leather instead of wet cotton.

Changing resolution up and down has similar effects plus collision distances between water/cloth/geometry gets to big or if they get too small interpenetration of cloth/geometry gets even worse.

Thanks for the description. Did you forget to attach the scene?

Note that cloth must be dense enough for particles in it to not contain gaps, otherwise stuff will go through it. My guess is that your cloth particles are small and thus other particles just pass through the gaps. More info:

Marsel Khadiyev (Software Developer, EPHERE Inc.)

Forgot to click upload, sorry.

Here is the scene

Attached Files:

>Lucid-Test-small.max (1744896 bytes),

A few notes I see:

  1. The cloth needs to be subdivided more. If you preview it as particles you can see that particles are really spaced widely apart. They need to be tight together as I mentioned earlier.
  2. The drum mesh can be broken into a few separate meshes which would be more easily segmented into convex collision meshes. I think that'd help the algorithm.
  3. The walls of the drum mesh can be made thicker to help prevent the leaking particles. They appear to be tunneling through the walls at the sub-steps.
  4. Additionally to #3 you can increase the sub-steps to something like 30 and iterations to maybe 10 to allow more accuracy and help prevent tunneling further.

Marsel Khadiyev (Software Developer, EPHERE Inc.)