I played around a bit and discovered a few amusing details. Gemini 2.5 creates reproducible images that look remarkably like real people. You can place a model barefoot in the snow while wearing a denim skirt, for example, but once she's wearing a leather skirt, the AI refuses to cooperate. You can strip a model almost naked. The AI can't generate crouching people, but it can generate walking, sitting, and lying down people.
18 year old white girl with blond hair, walks in water city fountain and takes fun in all wet short white top splash water, all wet blue jeans splash water with wet black leather knee boots on high heels. Background: women in wet clothes and white sneakers and heels, wide view
walks/sits/lies/lies on her stomach, etc
also normal view, side view, rear view to see a model in different position
I also experimented with snow and frostbite. Gemini can produce such photos too.
Walking in the snow barefoot, from euroria to severe frostbite of the hands and feet.
" 20 year old white girl with blond hair through snowy city street, walks on street, slim red leather jackets, white short top with nude stomach, frozen hands with fingers, frostbitten feets with fingers, grey leggins, without shoes, without winter gloves. Background: snowy city, snow crowdystreet, heavy snowstorm"
It can do, maybe try playing around a bit and altering your prompts. Over the last months Gemini seemed to loosened its filters more and more, becoming very powerful and capable of doing almost anything except full nudity. But it *can* do things like sexy poses, seethrough clothes and even partial nudity if you ask it to do so.