A checklist for evaluating AI image generators when typography and layout matter

A checklist for evaluating AI image generators when typography and layout matter

Most AI image tools are judged on how pretty the output looks. If you are producing anything that carries text inside the image, that is the wrong test.

Here is the short checklist we use:

1. Render three words at the same size in one image. If the letter spacing drifts between them, the model is not ready for layouts with headings.
2. Ask for a two column layout with a clear gutter. Models that "cheat" by blurring the middle fail here.
3. Test the same prompt with a long body of text. Most tools degrade badly past a few words.
4. Check the negative space. If the model fills every corner, there is nowhere to place a caption later.
5. Re-run the winning prompt three times. Consistency between runs matters more than a single lucky result.

If you want to see how a generator handles motion instead of stills, the same consistency test applies to video: try genjutsu AI and check whether the subject holds its shape across the clip.

Nothing here is exotic, but it filters out most tools in under ten minutes.