AI Image Generator - Prompt Discussion
Posted: Sat Apr 20, 2024 6:23 pm
Hello everyone,
I recently came across this AI image generator, and I have found it very reliable in terms of creating photo realistic quality pictures of people, though it struggles to generate realistic nudity, I have been able to generate nudity, it just looks pretty rough.
So I thought I would open it up to the community, to discuss what prompts can be used and what settings people like in order to create characters for web teases.
Just from playing around with it so far, my typical settings look like this, even though I'm not sure what any of them actually do:
Prior Guidance Scale: 6-14 (this seems to scale up how hard the AI follows my prompt, though going too high makes things a bit abstract)
Prior Inference Steps: 10-26 (the lower this is the more variety I seem to get between images)
Decoder Inference Steps: 10-12
Typical prompts I type in at the moment to push the AI in the direction I want it to go in:
point-of-view from above (I find giving it an idea of where you want the camera to be dials it in on other prompts)
full-body (this generally gets me a more zoomed out photo focusing less on just the face)
Succubus (this doesn't always give the character horns, but the AI definitely shifts the general vibe of the image when this prompt is used, and you can put "horns" into the negative prompt to make sure you don't get horns if you don't want them)
in an armchair / on a bed / on the beach / on the dancefloor / at a bar (Giving the AI a location of any kind seems to help it stay away from producing abstract art images instead of an image of a person)
freckles/glitter (Forcing the AI to generate detailed features like freckles or glitter typically means the entire image with look more crisp and less fuzzy)
Cartoon & Animated (I put these in the negative prompts as sometimes the AI will go down a rabbit hole of creating cartoon images or abstract concepts instead of people)
dress/sweater/crop top/bikini (Giving the AI direction in the clothes worn is very effective at making sure you get a full body image and not some AI mess)
Post any prompts you figure out that generate some nice images!
https://huggingface.co/spaces/multimoda ... le-cascade
I recently came across this AI image generator, and I have found it very reliable in terms of creating photo realistic quality pictures of people, though it struggles to generate realistic nudity, I have been able to generate nudity, it just looks pretty rough.
So I thought I would open it up to the community, to discuss what prompts can be used and what settings people like in order to create characters for web teases.
Just from playing around with it so far, my typical settings look like this, even though I'm not sure what any of them actually do:
Prior Guidance Scale: 6-14 (this seems to scale up how hard the AI follows my prompt, though going too high makes things a bit abstract)
Prior Inference Steps: 10-26 (the lower this is the more variety I seem to get between images)
Decoder Inference Steps: 10-12
Typical prompts I type in at the moment to push the AI in the direction I want it to go in:
point-of-view from above (I find giving it an idea of where you want the camera to be dials it in on other prompts)
full-body (this generally gets me a more zoomed out photo focusing less on just the face)
Succubus (this doesn't always give the character horns, but the AI definitely shifts the general vibe of the image when this prompt is used, and you can put "horns" into the negative prompt to make sure you don't get horns if you don't want them)
in an armchair / on a bed / on the beach / on the dancefloor / at a bar (Giving the AI a location of any kind seems to help it stay away from producing abstract art images instead of an image of a person)
freckles/glitter (Forcing the AI to generate detailed features like freckles or glitter typically means the entire image with look more crisp and less fuzzy)
Cartoon & Animated (I put these in the negative prompts as sometimes the AI will go down a rabbit hole of creating cartoon images or abstract concepts instead of people)
dress/sweater/crop top/bikini (Giving the AI direction in the clothes worn is very effective at making sure you get a full body image and not some AI mess)
Post any prompts you figure out that generate some nice images!
https://huggingface.co/spaces/multimoda ... le-cascade