Under Armour ‘Rule Yourself’ featuring Stephen Curry, Misty Copeland and Jordan Spieth

Under Armour’s ‘Rule Yourself’ debut spot multiplies Stephen Curry, Misty Copeland, and Jordan Spieth into armies of athletes perfecting their craft. Directed by Wally Pfister, the commercial unfolds from solo shots into thousands of duplicates — multiple Currys dribbling two basketballs, Spieth catapulting golf balls skyward, an army of Copelands twirling in a giant circle — representing the thousands of training hours behind every world-class performance. With limited time with the talent, The Mill shot each athlete on a special five-camera horizontal rig across ten takes, then created CG doubles using Massive and motion capture to populate each environment. The campaign later added spots starring Tom Brady.

To bring those Massive digital doubles closer to camera, The Mill commissioned full 3D head scans of each athlete on set, captured by Travis Reinke at SCANable.

Duplication only works when the original is real

A frame filled with a thousand copies of one person is a strange kind of stress test. The audience is not comparing the copies to reality — they are comparing them to each other, and to the single real performer the sequence started with.

Any error in the source asset is therefore repeated a thousand times. A slightly wrong proportion becomes a pattern, and patterns are exactly what human vision is best at detecting.

Measured capture is what keeps the multiplication honest. Every duplicate inherits the same accurate geometry, so the crowd reads as many instances of one real person rather than many instances of an approximation.

The head is where a duplicate gives itself away

Bodies at distance are forgiving. Silhouette, posture and motion carry most of the read, and a moderately accurate body will pass in the middle ground.

Heads do not work that way. As soon as a duplicate approaches camera, attention goes straight to the face, and the tolerance collapses. Skull proportion, hairline, the relationship between eyes and nose — these are the features that identify a specific person, and they are checked involuntarily.

That is why the near-camera copies need capture at a different level of fidelity from the ones filling the back of frame. Our digital doubles work is scoped around exactly that distinction.

Crowd systems consume geometry, not photographs

Simulation software distributes and animates agents. It does not invent them. Whatever the system is given as source is what the audience ultimately sees, multiplied.

Feeding a crowd engine accurate scanned geometry means the agents already have correct volume, correct scale and a surface that responds properly to the scene’s lighting. Feeding it approximated stand-ins means every rendered agent inherits the same flaws.

Getting the source right is cheaper than fixing it at scale, which is the case for cyberscanning as the front end of a crowd pipeline rather than an optional extra.

Additional commercial projects is collected in the portfolio, next to the full service list.