AI image generators don't draw. They start with noise and gradually adjust the RGB values of each pixel based on patterns learned from billions of photos, like a sculptor chiseling away stone. Nick Perry explains how a text prompt guides this diffusion process to create images that never existed before.
Did you know that AI doesn't draw the images it makes of you? What it actually does will shock you because it's way weirder than that, and here's why. So, picture yourself as a sculptor and imagine someone asked you to make them a statue. What would you do? Would you add things to the stone, or would you chisel away everything that doesn't contribute to the shape of the statue? Yeah, well, AI works the same way except it uses pixels. Because a machine doesn't see a picture the same way you do. It sees a grid of pixels instead. And every pixel is represented by 3 signals: red, green, and blue. So, the way AI learned to make images of you was by seeing how the signals were arranged across billions of photos and replicating those patterns. See, this is called diffusion, and it's what happens when you type a prompt into AI image makers.
