Text in images is mostly a solved problem with frontier models. At least regarding consistency. Most of the time it's still obvious and looks strange in natural-looking "photos".

This is an example generated by Qwen Image as a single image file: