I have been trying to use AI for the last few days to get decent SW imagery and find that I can't really coax any of the free sites into giving me a way to change the scale of characters. Most of the time the AI just ignores my size prompt completely and when it doesn't the character just winds up distorted or looking like a plastic toy. For example...
Additionally I've been struggling to find a way to make a character consistent over multiple images, in order to make comics. In just about every case if I type the same prompt multiple times the character comes out dramatically different in each picture, often just ignoring keywords entirely (like when I say "black boots" and get white ones or no boots at all). I know you can trick Stable Diffusion into making a consistent face by telling it to combine celebrity faces but that doesn't help when the clothing is different every time and I need the same costume throughout.
One site that gives me very nice images and consistency is Katalist, but so far I've been unable to get the character to look anywhere other than directly at the camera, and also, making her smaller is nigh impossible. Every attempt to rescale the figure fails completely and prompts telling the AI to use giant scale backgrounds are ignored. If I could get past that, I could make some incredible comics. I mean, seriously, imagine what Sandi's Trials would look like if I could create it using something like this...
By the way, the prompt for this image was:
#TinySandi (scale: 10%) shrunk down to four inches tall, wearing a white sleeveless leotard, white opera gloves, and white thigh boots, standing with legs apart, gaze looking up in horror as she stands on the Giant Tabletop, next to a gigantic book. Dramatic lighting, tense mood
Notice how the "gaze looking up in horror" was completely ignored (Katalist pays no attention to any instruction that would alter the character's gaze to any direction other than directly at the camera, and the posing controls don't do anything for gaze direction) and even defining a location like Giant Tabletop and saying "gigantic book" does nothing whatsoever to alter her scale, just like "scale 10%" and "shrunk down to four inches tall" are ignored. Once, in one render, I got the AI to produce a duplicate image of her at half normal size, but it was an anomaly that never repeated. *sigh*