~2 minutes of gaming per card
Every image in Mortal Arcana is AI-generated, created locally and on-prem on a single mid-range NVIDIA RTX 4060. No cloud render farm, no hyperscale data center — just the electricity it takes to run one desktop. A 4060 pulls the same ~115W rendering a card (~2 min) as it does running a game, so each card is, in energy, about 2minutes of someone’s favorite shooter. That’s the honest unit.
On-prem, not a data center.
One mid-range consumer GPU in a room, drawing the power of a desktop. There is no rack, no campus, no cooling tower. Just a tower under a desk.
We run, we don’t train.
We use pre-trained open models plus small local LoRAs — none of the giant foundation-model training cost. Our footprint is just inference: the few seconds of compute it takes to draw one card.
We count everything.
The tally includes drafts, rerolls, and finals — every render that ever drew power. Nothing hidden behind “final cards only.” The number up top is the whole receipt.
The honest caveat
Those measured numbers cover our energy — the watts to draw each card. They do not include the upstream cost of training the base models we build on, which happened before us and which no one running inference can fully account for. We count what we burn; we can’t count what we didn’t.