Meet Judge Bob, the AI judge
Somebody has to decide who won, and a vote would just be a popularity contest. So it’s an AI! Judge Bob gets every drawing the moment the clock stops, scores each one on five things and names a winner with one line on why. He sees the pictures and the prompt and nothing else. He gets it wrong sometimes. This page is what he’s looking for.
What Judge Bob scores
Four craft axes score up to 100 each, for 400 total. Accuracy then multiplies that by 0 to 1. Draw the wrong thing and the rest barely counts!
Accuracy
Did you draw what the prompt asked for? This multiplies everything else, so a gorgeous cat scores almost nothing when the word was "lighthouse". Writing the word instead of drawing it scores very low. He checks!
Creativity
What you did with the prompt beyond the literal thing. When both drawings are clearly correct, this is usually what decides it. Give the lighthouse a storm!
Linework
Proportion, control of the line and closed shapes. A crude or silly style still scores here if it’s drawn with control.
Color
Whether the ink does any work. Black and white scores full marks when the grays are deliberate. One flat ink over everything loses points, and a paint-bucket background counts for nothing.
Detail
How much is in the subject itself. A cookie drawn as a circle with dots scores low. Give it an uneven edge, some crumb and different-sized chips and it scores high! Scribble doesn’t count.
Is that fair, and can I argue?
Every drawing goes to Judge Bob together, labeled only as seats, in an order that says nothing about who drew what. He isn’t given a name or a rating. Order matters more than you’d think: shown the same two drawings, the model gave whichever it saw second an extra 40 to 100 points out of 400. So the order now comes from the drawings themselves, and it can’t favor whoever joined first.
The verdict is final. Instead of an appeal you get the score on each axis and his one-line reason. He’s software and he’s wrong sometimes. When you think he is, queue again.