How to keep consistent characters in prompts
I have a question about how to make consistent characters in a prompt. I want to make two specific characters interacting. My example prompt is below. But the generator keeps mixing and matching the appearance of the two characters. I specify different hair or clothing colors for each, but sometimes both characters have the same, mixed, or reversed colors.
How can I write the prompt better?
Charlotte (slim adult human woman, grey jumpsuit, long blue hair, smug smile) is laughing at Kristin (adult human woman, green dress, short blonde hair, embarrassed blush)
The generator doesn't know Charlotte or Kristin, there is no information about the location or anything at all. You're just putting two vaguely different piles of keywords into a void, leaving the gaps for the generator to fill in.
Here are two quick & dirty examples of how you could do it (graciously provided by an AI prompt enhancer helper character). You don't have to do it exactly like this, you can use art styles etc. the generator you are using provides, you can leave out details for the generator to fill in, but you should give a clear idea of how things/persons interact with each other so it actually "knows" what you want. Notice how Charlotte is always on the left and always looks like her description? And how there's always the bench and the fountain, the park, leaves etc.?
Charlotte, a slim woman in a practical grey jumpsuit, leans casually against a wooden park bench on the left side of the frame. Her long blue hair spills over one shoulder as she laughs with genuine amusement, eyes crinkling at the corners while pointing playfully toward Kristin. Sunlight streams through golden autumn leaves, illuminating dust motes dancing in the air around her. Kristin stands near a mossy stone fountain on the right, fiddling with the hem of her green dress as she stares at rippling water. Her short blonde hair catches the afternoon glow, cheeks flushed pink beneath freckles visible in the natural light.
Charlotte, a slim woman in a grey jumpsuit with cascading blue hair, leans against an ornate park bench on the left side of the frame. She laughs openly, one hand covering her mouth while the other gestures mockingly toward Kristin. Sunlight filters through autumn oak leaves, casting dappled shadows on her smug expression. Kristin stands near a fountain on the right, her green dress rumpled as she stares at her own reflection in the water. Her short blonde hair catches the golden hour light, cheeks flushed pink as she nervously twists a fallen maple leaf between her fingers.
---
so perchance does understand the sentence structure you're using. I was using the parentheses as more of a piping command, as it was advised the parenthesis could make the generator pay more attention to specific terms. I'll try your suggestion!
That was correct for the previous Stable Diffusion model, but no longer applies to the current customized Flux.1-schnell
And yes, I've asked on the Casual Perchance forum as well. Just figured there would be a bit more expertise in this one!
One thing that you can do (now bear with me because this might take a bit), is to go to the ai character generator, keep prompting until you find an image that you like, then chat with them and then whenever you want to make a specific scene, describe it in the chat and then click the image button. It's a very complicated process and i wish there was a more direct way, but so far i haven't found anything yet, so this is the next best thing. (link for ai character generator https://perchance.org/ai-character-generator)
I'll give that a try. Tutorials and posts about other prompts, line breaks, arrays, etc aren't working. So a long process might still be better than "keep pressing the button until I don't hate it." 😉
huh... is this because it takes the seed of the image you are chatting with to render new images? The textual persistence is so off the wall that I haven't bothered trying to create cohesive or connected visual pieces or scenes.