We bet almost two decades ago that data was the thing that moves games forward.
Game development sits at the crossroads of technology, creativity and art. It is also, increasingly, a data problem.
Kapnetix started in 2020 as a research project for one of the world’s largest game publishers, who wanted to know what AI could realistically do inside game development. We expected to come back with a list of opportunities. We came back fixated on one.
The number that did it: one hour of raw captured motion takes roughly one day to shoot and five days to clean up. Stage time runs well over $10,000 a day, and the expensive half is not the stage. It is the week afterwards, spent by animators opening files one at a time to find the handful that will cost them the schedule.
So we went to the field rather than the whiteboard. GDC, Game Access, capture stages across Europe, and a lot of conversations with the people who actually do cleanup. The same answer kept coming back, and it was never “make it more accurate”. It was “tell me which takes I have to open by hand”.
To build that, we needed data nobody had: capture with known, deliberate errors in it. We commissioned it. At eNStudios in Poland we ran an OptiTrack volume and asked performers to produce the exact failures that ruin a take, then repeated the exercise across ten studios. That is what our models learned from, and it is why we can say your takes never feed them.
The name is capture with neural network kinetics. The mark was drawn by a founder’s seven-year-old daughter, and the coloured dots joined by gradient lines turned out to be exactly what the product renders. We did not plan that. We kept it.