The idea
[ SEC 01 / 04 ]Someone walks up to a screen and asks a question — out loud or by typing — and a life-size, photorealistic figure answers aloud, in character, with its face and body moving in time with its voice.
Built as personal R&D around one demanding brief: it has to run all day in a public space, with no operator and nobody there to restart it.
The challenge
[ SEC 02 / 04 ]-
01
Feel instant
A silent, staring figure reads as broken. It starts speaking in about two seconds — quick enough that the pause reads as thinking, not failure.
-
02
Stay truthful
It answers from a curated body of knowledge, and when it doesn't know, it says so. An honest "I don't know" beats a confident wrong answer.
-
03
Never break in public
No error screens, no frozen faces. If anything behind the scenes stops, the figure settles calmly and the visitor sees nothing alarming.
What made it hard
[ SEC 03 / 04 ]A digital human looks alive in the small moments, not the big ones — a blink that lands naturally, gestures that fit the length of a sentence, a body that keeps breathing while it listens. Each was tuned by watching the character answer real questions.
The hardest problems disguised themselves as something else: what looked like one fault was usually two, quietly cancelling each other out. They were solved by measuring what actually happened, not by trusting the most plausible explanation.
The result
[ SEC 04 / 04 ]A self-running kiosk that holds a spoken conversation in character, keeps its composure when something goes wrong, and needs no one to look after it — built end to end, from the character to the screen the visitor sees.
Happy to walk through it in person.