FX TD

What they do

An FX TD gets a shot with approved animation and a note like: the wave has to break over the boat on frame 60, and the character has to be visible through it. What they deliver is caches, usually volumes and geometry sequences written to disk, that lighting and compositing render. The work is setup, not painting: building a network in Houdini, choosing a solver, wiring collisions to the animated character, and then running the simulation over and over at increasing resolution while inspecting whether it does the thing the shot needs. Iteration is slow and expensive because sims run on the farm, and a low-resolution test that behaves does not guarantee the high-resolution version will. Art direction is the hard part. Physical accuracy is available and frequently wrong: published work on Disney's water for Moana is about making a simulated ocean perform, hitting story beats, holding a shape, acting like a character, which no unguided solver will do. So FX TDs build layered setups where a hand-animated guide drives a sim, and cheats sit under the physics. The role is also a tools role. FX TDs write VEX and Python, wrap setups so other artists can use them, and are the people who explain why a cache is 400 gigabytes. Character FX, the cloth and hair on every character in every shot, is quieter, higher-volume work in the same discipline.

What they decide

The calls that belong to this role and no other.

• Which solver and approach an effect uses, and whether it is simulated, procedural, or hand-animated
• The simulation resolution and substep settings, traded against farm time and disk
• Where physical accuracy is abandoned for art direction, recommended to the supervisor
• How an effect is layered: guide animation, base sim, secondary sim, whitewater or wisps
• What gets cached and in what format, and the storage cost that comes with it
• Whether an element is delivered as a 3D cache or handed to compositing as a 2D solution

Principles of the craft

• Art-direct the sim, do not referee it: a solver left to be correct will produce water that does something plausible and nothing dramatic. Build a guide, whether that is animated geometry, a velocity field, or a hand-shaped volume, and let the physics add detail on top of a shape you chose. Disney's Moana water work is a published example of exactly this approach.
• Layer effects the way the eye reads them: a wave is a shape, plus foam, plus spray, plus mist, and each layer has different timing and a different failure mode. Simulating them as one giant expensive sim gives you no way to fix the one that is wrong.
• Test at low resolution, but know which behaviours do not survive scaling: bulk motion and timing hold as you increase resolution, while fine detail, splash count and surface tension change. Learn which of your notes can be addressed at test resolution and which have to wait for a full run.
• The cache is the deliverable, so budget it: effects work is measured in terabytes and farm hours as much as in frames. Deciding what resolution the shot actually needs, and what can be an instanced or procedural cheat, is a creative decision with a line in the schedule.
• Wedge, do not guess: run the same setup across a range of one parameter and look at the results side by side. Physically-based systems have too many coupled controls to reason about by intuition, and a wedge turns a day of blind tweaking into an afternoon of choosing.

Common mistakes

• Trusting the solver to be interesting, and delivering technically correct fire that has no shape and hits no beat.
• Jumping to high resolution before the low-resolution behaviour is right, then discovering at hour nine that the timing was wrong all along.
• Simulating a solution that could have been a hand-animated element or a 2D pass in compositing at a fraction of the cost.
• Delivering caches that lighting cannot render: motion blur data missing, no velocity fields, or sizes that break the farm's memory budget.
• The experienced TD's mistake: building an elaborate general-purpose setup for a one-shot effect, when a specific cheap cheat would have shipped days earlier.

How to learn it

Exercises and routes in, not a reading list.

• Install Houdini Apprentice and work through the official documentation's pyro and FLIP examples, then make one shot: a mug of liquid poured out, hitting a specific mark on a specific frame. You will discover how much guiding is needed to hit a frame at all.
• Wedge one parameter, viscosity or particle separation, across five values on the same simulation and lay the results side by side. Write down what changed and what did not. This is the single fastest way to learn a solver.
• Simulate the same effect twice at two resolutions and compare timing and bulk motion. Note which of your creative notes could have been judged from the cheap version.
• Read the Disney Animation Moana water publication, then set yourself the same problem: make a body of water perform a beat, arrive on a frame, and hold a readable silhouette. Do it with a guide first and physics second.
• Work through the Khan Academy Pixar in a Box effects and simulation units for the underlying particle and constraint ideas, then rebuild one of those simple systems from scratch in code or in VEX so you know what the solver is doing.