I am about to embark on a tricky cloudscape for a film that I am working on, and was wondering if anyone had any advice for using the cloud flows on large PRT volumes, or any tips on how to go about building up large cloud shots…
I have attached a concept of the scene I have, and the camera flys directly towards the mountain in the middle distance.
I have been testing out different scales for the scene and have smaller clouds looking great…problems with the big ones though…
Splitting the scene into multiple layers by distance and compositing the foreground, middle ground and background clouds might be the way to go. This way, you can use smaller voxel sizes to show more detail in the foreground, while not oversampling the background. The main issue will be when the mid-ground clouds become foreground as you fly closer - slow blending between two image sequences rendered with different settings might be required to slowly reveal more detail…
The right way to solve this case would be to implement frustum-aligned grids where the size of the voxels varies within the camera cone, so you get small voxels in the foreground and larger voxels in the background. Right now this is not part of the Krakatoa voxel renderer.
Another solution that was proposed recently was the ability to override the voxel size via the Krakatoa Camera Override modifier and allow it to be keyframed over time, so you can have a dynamically changing voxel size that gets smaller as you move closer to the foreground clouds layer. Once again, this is currently not possible in Krakatoa since the UI controls are not animatable. It can be hacked through some creative MAXScripting, but it is not straight-forward.
I’m not sure I’d use voxels on this. I’d probably start with a point render. There’s very little illumination interaction between the various cloud layers that can be seen. So it makes sense to break these up both in depth and in distance from the sun. You should be able to render out the upper clouds along with the shadows to be cast on the next layer down.
There’s isn’t multiscattering in Krakatoa either. So you can get away with cheating the density of the clouds a lot. They can be nearly hollow, provided the shadow map can pick up just a few super-dense, super-absorbing particles inside. Some of the shading effects in your image can’t be achieved at all in Krakatoa without some cheating, either, but that’s fine too. You can add fill lights or do projection textures or pre-light some mesh primatives that can transfer their illumination to the particles via a KCM.
After playing all day, going home and thinking about it since posting. I came to the conclusion, that cheating will be the way to go!
As I am not the most accomplished of coders, scripting trickery is not really an option for me, so I have to approach it from more of a compositors perspective, as suggested, splitting the scene into layers and sculpting one cloud at a time to the directors whims and wishes
My plan is to create varying scales of the scene, keeping the camera frustum the same between shots, and building up a ‘cloud set’ that uses the most suitable voxel spacing etc. for each ‘style’ of cloud.
The art team are blocking out the cloud shapes using simple geo, then I will go in and tweak the sculpts building out the form more, and create mini clouds that can be placed and added to increase the detail in the larger clouds.
Your suggestion Chad to try out particle rendering, plus using pre-lit meshes sounds great too, I will try that.
I am starting to get some really nice detail in the close up fly-over clouds, that will hopefully hold the shot together, and the distant clouds can be mattes, or projections in nuke onto some displaced geo…