AI can paint a face so real you'd swipe right on it, then spell SALE as SLAE on a poster. Canva just fixed that by making its model plan the poster like a designer before it paints a single pixel. Many image models go straight from prompt to pixels in one pass. To them, the word SALE is closer to texture, painted at the same moment as the brick wall behind it. I make a lot of my own visuals, and one wrong letter has meant re-rolling the whole image and losing the parts I liked more times than…