Matters of form, scale, and layout
I liked the L’il Esils so much that I decided to make more, adding the challenge of putting them to bed without the model turning it into loli porn. Careful use of negative prompts managed to produce usually innocent pictures of L’il Esil with innocent sleepwear, stuffed animals, and non-sexy poses, but the attempt exposed another common problem with generative AI: it doesn’t understand how rooms work or how people fit inside them.
The basic prompt was: “<lora-trigger> wearing black and red frilly pajamas and fuzzy slippers. sfw, happy, flat chest, child height, child body, child face. indoors, girl’s bedroom at night, evil stuffed animals in background” (and, yes, despite the lack of nudity and sex keywords, I had to add “cum” to the negative prompt; in one pic, the teddy bear had a puddle of white stuff leaking from its crotch, sigh).
A human being would interpret “child height” in relationship to the typical sizes of common bedroom objects, like doors, chairs, dressers, and beds. Generative AI sometimes seems to get it right, but it’s really the user deciding between equally-likely alternatives.
So for fun, here’s a whole bunch of L’il Esils, and the challenge is to notice how many things are not-quite-right about the images.
Comments via Isso
Markdown formatting and simple HTML accepted.
Sometimes you have to double-click to enter text in the form (interaction between Isso and Bootstrap?). Tab is more reliable.