Search the Community
Showing results for tags 'sparse upres'.
Found 1 result
Sparse Pyro Upres H18
k2p8 posted a topic in Finished WorkHi everyone ! I have been working on a sparse pyro upres recently and did a video to test it. You can find the hda here : https://drive.google.com/drive/folders/1H4Y0HNELDJ6lmHsM43eEI_lRgNuCEaxX?usp=sharing More details about the solver : The sparse upres solver is a deeply modified sparse pyro that allow the user to quickly upres a smoke simulation from simple inputs. The core idea for the upres is to bypass every nodes related to velocity update in the smoke solver, like the project non divergent, vel advection, collision mask, ect. and replace the upres vel field by a copy of the low resolution velocity. One can then advect any dyes needed through that vel field to create an upres. In order to keep the pipeline as simple as possible i splitted the process in 2 nodes, a 'configure_inputs' which precompute some lowres fields and normalize inputs and the upres solver itself. The sparse solver needs to match the parameters of the lowres simulation to work properly (like dissipation, flame amount, temperature diffusion, ect.). To quickly set up the upres solver i added a python script, that copy and paste all relevant parameters of a lowres pyro to the upres solver. You just need to select which pyro solver to use as reference to match the parameters. The sparse solver itself has some simples parameters like noise / random that allow the user to inject details in the lowres vel field. Noises and randomness are stored on an separated field, 'noise', to be conserved through time since vel is simply copied from the lowres. That allow the noise field to be advected by the vel and then merge with it to advect other fields like density, flames, ect. Which create a more complex pattern than a noise directly merge with the vel. Color can be added at the upres stage if the source contains a Cd and Alpha volumes. Those fields will simply be advected by the upres solver like any other fields resulting in a colored smoke. Your lowres simulation needs no color. Retime can be perform by interpolating the source and scaling parts of the upres solver. The process is automated from a tab in the sparse solver. However, it can vary if you animate the retime or set it to low (especially with flame). The upres solver does not need any modification of the velocity other than small disturbance / turbulance details. If you add strong changes in the vel field your upres dye will deviate far away from the lowres vel field. Every major modifications of the vel field like wind or buoyancy need to be added in the lowres simulation. A fake divergence can be added through the process by computing the negative gradient of the density added to the velocity. Since pressure projection has been bypassed, important amount of this 'divergence' will result in compression in the vel field creating artefacts in the smoke. However, it can be usefull to break the sharp edges of an upres smoke if needed. At last, one can force the upres to match the lowres fields by simply removing parts of the simulation which are outside the lowres smoke by taking the lowres density field as reference. If the lowres is below a threshold value, the upres smoke is multiply by amount less than 1. This can be usefull is you have strong noises and need to dissipate them to stay in the bounds of the lowres.