Adult images were not part of the training data and a censoring mechanism is built into image creation. If the result seems too ‘naughty’ it will be blocked.
3 Likes
I was eagerly anticipating being able to play with this stuff, and then I did for a while, but now I’m kinda bored of it. Maybe I’m just not imaginative enough. Or cause I don’t get much out of a photo once I do make it.
I think I’ll cancel my Midjourney subscription which I haven’t used in six weeks, and buy some Dalle credits to play with the outpainting and inpainting a bit (editing and expanding existing photos)
1 Like
Those are amazing. Are those from stable diffusion?
Yes. I did not make those, but here are the details:
Using AUTOMATIC111’s interface.
a vogue portrait of a happy xenomorph alien and a xenomorph alien, at a bar, Having a drink, Laughing, artstation, cgsociety, very detailed, intricate, detailed illustration, by artgerm and greg rutkowski and alphonse mucha, octane render, unreal engine, hyperrealism
Negative prompt: (humans), photoshop, render, men, man, women, woman, video game, 3d, painting, art, drawing, digital art, cartoon,
Steps: 150, Sampler: LMS, CFG scale: 7, Seed: 3768044754, Size: 896x512
3 Likes
Can it do Rigellians? Craiyon and whichever other one I tried didn’t understand. Even “one-eyed green alien with tentacles” was a disaster, especially when I added “hippie”.
1 Like
Try space octopus.
1 Like
You have to use one that allows “Negative prompt”. You can exclude things to improve the result.
There are also methods to add your own pictures to training set to achieve certain things or style. But you need a decent computer with lots of v-ram to train the model (ca. 6h+).
2 Likes
I prefer the one in red. Is that the evil clone?
But… If one is already bad, then the evil clone would be good, no? 
2 Likes
What previous thread?
I think there’s only this one.
This is the previous one. It was merged already.
1 Like
1 Like