> How do you get the AI do Lolis?

One thing that sometimes worked, to my suprise, was simply to decribe a character as "a petite very flat-chested woman". I could sometimes say "girl", but "woman" worked better to avoid hitting roadblocks. That simple wording alone sometimes produced startlingly young looking characters.

What helped more was the way these text-to-video (and text-to-image) prompts can use images as input material. Speak as if the face of a girl you're providing (not even a real girl, could be a face obtained from the AI in a non-sexual context) is the face of a woman, and then request that this "woman" be shown completely naked.

Content filters won't stop you from producing clothed lolis or just the faces of lolis, so such things can be starting points.

Once you've got an image of a naked loli via the above tricks, that image can then be used as input for a more detailed scenario.

No guarantees any of these tricks will always work. Sometimes an input image gets rejected for looking to young (obviously not very consistently, given the results I managed).

Also, a bit randomly, these lolis can morph into looking more mature. Sometime it happens, sometimes it doesn't. Sometimes not at the start of a video, but only later in a video.


The best way to get an AI to do lolis is what I hope to have for myself soon - your own independent computer setup for text-to-image and text-to-video work.

I've got a more powerful graphics card arriving tomorrow, and hopefully after I install it I'll have much more success installing and running SwarmUI (an enviroment that supports ComfyUI, Stable Diffusion, etc.) than I did when I tried that with the much less capable graphics card I've been using.

You're the boss when you make your own setup. You can turn off all the content filters you want to turn off.