How do you write AI image-generation prompts?

Don't just say 'make me a picture' or 'make a poster'; that only produces cookie-cutter AI designs. The author suggests writing more specific prompts instead.

Start with the layout: the Danhuangbao example sets a 4:3 landscape format, with about 58% on the left for the original photo and about 42% on the right for the pattern; Xiaoxiaodong's example is a 3:4 portrait format, with 50% each for top and bottom.

When making an image from an uploaded photo, state what to keep: subject structure, real texture, natural lighting and shadows, and original colors. You may naturally extend the background, but do not stretch, distort, or alter the subject.

When drawing people, spell out each detail: the Grayscale Notes example specifies frame size, photographic texture, about 24 years old, skin tone, and hairstyle, and assigns a separate color to each of the four qipaos so the four people are clearly distinguishable.

Include tone and material too: one person specifically requested a Morandi palette, and another specified an off-white aged paper background with fibers and grain, leaving large areas of empty space.

The same prompt can be tried on different models: Xiaoxiaodong says Grok also works well with his prompt set; another person found that newer image models can recreate a familiar old style with one sentence.