Text rendering in AI image generation is notoriously one of the hardest problems to solve. Letters blur, words scramble, and even simple labels turn into typographic chaos. So which model handles real, readable words on an image better? We put Seedream 5.0 Lite and Nano Banana 2 - both