The challenge

TEX is an animation orchestration system for the gaming industry. It is an evolution of the CBA system that was originally developed for HABTIC, and later DELPHIC. A single compiled library capable of assembling animation streams for entire crowds of game characters with organic, natural, repetition free results? Sure. It was a slow afternoon.

The problems

The problem here is singular and seemingly unsurmountable: full body animation files are large and extremely labour-intensive to produce. A simple walk cycle for a small character can be dozens of megabytes. Full body motion for a complex character performing a variety of motions is simply more than a mobile can easily hold, more than can be downloaded comfortably to any device.

And that's per minute, per character.

There currently isn't a technology capable of solving the issue of animating a large group of people independently, individually and for a long time without deploying the kinds of data amounts that only video games have at their disposal.

Generative models are only useful for small, unchanging animation skeletons, and need to be retrained on every different Inverse Kinematic chain of every new model. There was no easy way to deliver on something like TEX without breaking new ground both in terms of both workflow and technology. So we did.

The solution

After several experiments into generative techniques failed, the solution was a mixture of workflow enhancements and precision-engineered AI to deliver what was required, that seemingly impossible mission brief.

The solution in this case relied on cheating, hard. TEX works with additive animation blocks. The character's skeleton is divided in logical chunks and each is animated by at least one concurrent animation stream pieced together by looping blocks. Most animation loops are 3-5 seconds long but there is no limit either way.

By recombining these blocks based on the instructions of a multi-armed context engine, TEX is capable of efficiently assigning unique, non-repeating animation streams to as many characters as required, making each respond differently and uniquely to the same set of circumstances.

Some characters will respond to the cold, some to the stink, others will simply look around curiously based on environmental cues. And TEX orchestrates each stream for each character driven by context input and character personality. The basic TEX demo animates 100 characters walking around a city with four distinct environmental conditions for hours, all based on a single 140 second animation library.

The takeaways

TEX can perform a seemingly impossible task by side stepping the strict requirements of pure technology. Generating animation for dozens of characters that still exhibit humanity and personality is an overwhelming challenge for trained models.

By taking a different approach and intelligently recombining handmade animation blocks, TEX provides a force multiplier in animation production and playback that is simply unmatched, and unbelievable to those coming from traditional animation backgrounds. TEX is Mundane through and through.