Improving AI Image Generation to Reflect Diverse Ages, Body Types, and Abilities

The AI image generation skews VERY heavily to younger people who are all slim and quite often white unless I get really, really specific about what including people of different ages. So I end up with young or folks with grey hair. There is a spectrum of people between those two extremes. I would love for it to more naturally generate pictures that are more diverse in nature. Pictures that include different body shapes, sizes, skin color etc. The only person with a disability that I am able to generate is someone using a wheelchair which is clearly not someone who uses it regularly as they are all hospital wheelchairs, not the style that ordinary people use. And, asking it to generate a visually impaired person with a cane or using a guide dog results in some interesting outcomes. None of which are usable. Same with asking to have people using sign language. Can we train the AI to have a wide range of people? Thanks.