Downpour is an internal effects study. Six shots, no story, no client — a deliberately narrow test of one question: at what point does a full simulation stop earning its render time against a cheaper approach that reads the same on screen?
The brief we set ourselves
Rain falling on and running off a collapsing concrete structure. Six shots at increasing complexity, from a locked-off wide with rain and standing water, to a moving camera through active structural failure with debris and secondary splash.
Each shot had to be built twice where it was practical: once as a full simulation, once with the cheapest approach that could plausibly sell it. Then watched side by side at speed, in the cut, rather than paused.
What came out of it
The honest answer is: most of it does not need to be simulated, and the exceptions are predictable.
- Rain in the air never justified a sim in any shot. Instanced cards with good motion blur and a proper falloff read identically at 24fps and cost effectively nothing. Every frame spent simulating airborne rain was wasted.
- Rain interacting with a surface is the opposite. Splash, run-off, pooling, the way water finds an edge and sheets off it — that is where the eye goes, and a cheat reads as a cheat immediately. This is worth the sim every time.
- Debris splits by scale. Large pieces need real dynamics because their rotation and contact behaviour is what sells the weight. Small debris does not; it is over too fast to inspect.
- Secondary splash on debris impact was the biggest surprise. Cheap in the sim, enormously effective, and the single thing that most improved the shots that were not working.
Simulate what the camera lingers on and what touches something. Fake what is in the air and what is over in three frames.
How it was built
Shot 3 was taken to near-final before the other five were started. That is standard practice on any effects sequence and it is standard for a reason: it settles the look while there is one shot to change rather than six.
Sims were cached at final resolution with every version kept. On shot 5 a note arrived that changed the collapse timing, and the earlier cache meant re-running one layer rather than the whole shot. That is not a clever technique, it is just discipline, and it is the difference between a note costing an hour and a note costing a day.
Comps ran from 11 layers on the simplest shot to 24 on the most complex, and every shot was reviewed in the assembled sequence rather than on its own. A shot that works in isolation and does not work in the cut is a shot that does not work.
What it is for
When someone sends us a cut to bid on, the shot bands in our quote come from having actually built these comparisons. We can say which shots need a sim and which do not, and give a reason rather than a rate.
It is also, plainly, the piece we show when someone asks what our effects work looks like. A new studio has to make something before it can show something, and this is us doing that rather than describing work we have not done.