Algorithmic simulation of millions of individual elements — rain, snow, dust, explosions. Compute-heavy but the only practical solution for organic mass effects.
On set or in compositing, you will eventually need millions of identical or similar objects that move according to physical rules — rain, snow, dust, sparks from an explosion, flocks of birds. The brute-force method of keyframing or animating each individual object fails immediately. This is where the particle system comes in: an algorithm manages thousands to hundreds of millions of tiny primitives — mostly simple geometries or sprites — and applies rules to them: velocity, acceleration, lifespan, collision, force fields. Each particle follows a predefined logic, not an individual instruction.
How it works in practice: You define an emitter — a source from which particles are born. Then you set parameters: emission rate (particles per frame), initial direction and spread, gravity, wind, drag. Each particle lives for a certain time, can change color and size, and can interact with collision objects. Modern engines like Houdini, Maya nCloth, or even game engines render this in real-time or as a simulation that is cached. A single explosion can contain 5–50 million particles. The computational effort is considerable, but without particle systems, you would be managing millions of individual objects manually — technically impossible.
Classic applications: Fire and smoke work almost exclusively through particle simulation, combined with fluid dynamics (voxel-based flow). Rain and snow are the simplest cases — constant emitter, linear fall, drag factor. Explosions require multi-layered systems: fast, bright particles for the impulse, slower ones for debris and smoke behind. Sand or dust clouds benefit from wind fields and gravity manipulation. Each software has its own tools: Nuke (3D Gizmos), Cinema 4D (Thinking Particles), Blender (Cycles Integration), RenderMan (for final production rendering).
The trick in compositing: Particle output is usually a raw layer — velocity, depth, object ID as separate passes. This allows you to tweak motion blur, depth of field, or color corrections afterward without repeating the entire simulation. Cache is your best friend: simulations run through once, are buffered as geo sequences or bgeo files, then read and rendered as many times as needed. You only recognize errors in the final composite — therefore, make early previews. Ruining a take with incorrect particles is expensive.