Rabbit_Games
Vyemm Raider
- 1,870
- 4,241
I don't play with them because they all seem to cost, and I'm not that interested yet. But... can you add "vector image format" to the parameters?
AI Image Generation
- Thread starter Captain Suave
- Start date
Pasteton
Blackwing Lair Raider
- 2,953
- 2,135
is there an app or method to make ai take a base image and reinterpret it into something completely different but using the same base image? Not sure how to describe what I’m thinking. But for example say you have I dunno a pic of a chick with 5 dicks around her blasting her face; is there a way to make ai take that base image and turn it into something food based , or turn it into a nature scene or landscape shot, or a cosmic dust image etc etc; yet you can still ‘see’ the bukkake foundation?
does that make sense, having a hard time explaining what I’m thinking of
does that make sense, having a hard time explaining what I’m thinking of
Kharzette
Watcher of Overs
- 5,679
- 5,666
Pixels are a big part of how everything works because of convolution. GPU number crunching has historically always been about texture mapping. Everything works in small square chunks to cache well etc.
Even when I was fiddling with audio AI a few years back, audio was converted into mel spectrograms to get it into a pixely format.
If you look at the theory of how neural nets work, a cloud of disparate points should be way simpler and almost ideal, but all of the great chain of libraries all works off pixels.
BTW, if you want to take a source image and use it to create something new, stable diffusion's img2img is really good for that. The main factor to adjust is denoising I think?
I don't have it open right now, but around 0.25 is where you set it if you have an image you want slightly modified. Like if you have a good image you just want to upscale with a small amount of detail added.
0.5 would modify the image quite a lot. You could change a subject's outfit or hair color or the background. Anything above 0.5 would probably just create a whole new image based on the prompt.
Even when I was fiddling with audio AI a few years back, audio was converted into mel spectrograms to get it into a pixely format.
If you look at the theory of how neural nets work, a cloud of disparate points should be way simpler and almost ideal, but all of the great chain of libraries all works off pixels.
BTW, if you want to take a source image and use it to create something new, stable diffusion's img2img is really good for that. The main factor to adjust is denoising I think?
I don't have it open right now, but around 0.25 is where you set it if you have an image you want slightly modified. Like if you have a good image you just want to upscale with a small amount of detail added.
0.5 would modify the image quite a lot. You could change a subject's outfit or hair color or the background. Anything above 0.5 would probably just create a whole new image based on the prompt.
Tuco
I got Tuco'd!
- 49,736
- 90,122
Tmac
Adventurer
- 10,346
- 18,135
I see really cool implications for DND campaigns. DM’s can just tell a prompt what they’re looking for, iterate, and have an entire 3D campaign mapped out for their peeps.
Tmac
Adventurer
- 10,346
- 18,135
You gotta photoshop good hands over the shitty hands and rerun it I think. Or does it fuck up good hands and make them shitty?
Tmac
Adventurer
- 10,346
- 18,135
- 31,813
- 24,626
I forget, this got boring as soon as I figured it out.
Tmac
Adventurer
- 10,346
- 18,135
Kharzette
Watcher of Overs
- 5,679
- 5,666
Hmm the pure realistic model I used was Ares I think. I'm usually doing fantasy stuff so I'll mix in a weeby model as well for my stuff.
Just looking at my dir I guess I mainly use Chillout, Meina, there's one here called Henmixreal but I don't remember using it. I think I might have downloaded it to try it and then frogot about it. Also gape60 is a very old model made for dirty purposes, but I discovered that it is fantastic at doing outfits.
I've always wanted to try Loras but I've never been able to figure out how they work.
For prompts I have the most luck putting emphasis on the eyes. Without it they look weird, and it often zooms in a bit hopefully putting the hands out of view so I don't have to worry about fixing 9 fingered hands
Something like highly detailed eyes.
I've had almost no luck at all trying to shape elements of the eye. If you have enough anime in your model, sometimes you can get a heart shaped iris, but spirals and stars and such are really hard to achieve.
Just looking at my dir I guess I mainly use Chillout, Meina, there's one here called Henmixreal but I don't remember using it. I think I might have downloaded it to try it and then frogot about it. Also gape60 is a very old model made for dirty purposes, but I discovered that it is fantastic at doing outfits.
I've always wanted to try Loras but I've never been able to figure out how they work.
For prompts I have the most luck putting emphasis on the eyes. Without it they look weird, and it often zooms in a bit hopefully putting the hands out of view so I don't have to worry about fixing 9 fingered hands
Something like highly detailed eyes.I've had almost no luck at all trying to shape elements of the eye. If you have enough anime in your model, sometimes you can get a heart shaped iris, but spirals and stars and such are really hard to achieve.
- 31,813
- 24,626