First of all, very nice work! Thank you for sharing. Loved the one with the giant apple so much, I hope you don't mind that I've decided to create a few variations myself. Also loved the one on the scale, drinking the shrinking potion, very creative!
Allright, I have to go to work but quickly:
rscholar wrote: ↑Mon Nov 06, 2023 4:50 am
-The censor mostly refuses to render prompts bare feet, as in "woman standing next to giant foot."
In my experience, rendering shoewear is actually harder than bare feet.
It really seems to be the default state. Have you tried simply not mentionning it? You have to get lucky, no doubt, but using "standing next to normal sized woman" can work if used in conjunction with an "extreme low angle" / view from the ground and "zoom in on tiny woman" to make sure to POV is close enough.
I just tried mentionning "bare foot" in a quick prompt that gave me two nice renders after triggering the fail dog 5 times in a row (sometimes you have to try 10 times in a row). Here's results from "
Create an image of a very tiny, shrunken wet young woman, standing in a field of oversized grass with morning dew, curiously looking at a bare foot that is taller than her, from a low angle perspective, with ambient sunlight filtering through the grass, in a hyper-realistic style"
NOTE: These are raw outputs, of course I would edit out the extra toe, do another pass in Stable Diffusion to clean up the faces and upscale the resolution.
rscholar wrote: ↑Mon Nov 06, 2023 4:50 am
-The censor also has a problem with "swimmer" or "swimsuit" - I can mostly understand it, but I once typed "female athlete" and got a woman in a sports bra and panties so I don't know.
Likewise, I had little success with describing clothes (or lack thereof), I don't have an exact formula but something to remember is the "mood" of the picture, the context if you will. Where do people wear swimsuits? Pool, beaches, etc. Craft that in your prompt, even if it's nonsensical. Another thing to remember is that the emotions in your prompt can help get by the filter, ie. positive moods, happiness, smiling, etc. can actually help.
rscholar wrote: ↑Mon Nov 06, 2023 4:50 am
-It's really tricky to render different sized characters and environments so a lot of my successes are mainly a woman next to an item and the occasional lucky strike. I try things like "woman next to giant/enormous item", but anything with another character or describing environments just gets jumbled. Worse, describing a normal person and a "small/tiny" woman ends with the woman as chibi or childish instead of proportionately shrunk.
This is the part where my info is mostly contextual, meaning that it highly depends on the entire prompt you crafted. Personaly I found that using "giant" does indeed scale the object and works well in simple renders, but in a complex scene it'll break the size illusion because it'll keep the rest of the scene at normal scale. I had better success using "very tiny, shrunken woman, only a couple of inches tall" in my prompts, and then specifying normal-sized object and scene size relative to her ("dwarfed by a bottle", "blades of grass tower around her like trees", etc). Always give something for the AI to latch onto for the size difference, and mention it several times in the prompt ("woman is proportionately minuscule compared to the hand's lines and contours").
I have to run but don't hesitate if you have more questions... Oh, and here's my tribute to your wonderful apple picture - raw output, they would need to be cleaned up and tweaked, of course.
