Lord Caramac the Clueless, KSC
@modean987 I've been using all kinds of AI tools for images ever since #DeepDream happened in the 2010s, that thing with the puppyslugs. You input one image, select which neuron layer of the model you want to sample, and how many iterations you want, and you get a new image that somehow grew out of the old one. A little later, Neural Style Transfer aka #DeepStyle made it possible to mix two images, one for content and one for style. I made enormous volumes of images with that. #aiart
Mar 14, 2024 at 6:50pmWeb
Show threadLord Caramac the Clueless, KSCMar 14, 2024
@modean987 After that we had all kinds of GANs to generate images, mostly very weird and often disturbing, some outright body horror, but I really like images like that. And then latent diffusion was discovered, yet I never used it when it was new because it was only available in the pay-for-play application Dall-E (1.0 back then). Stability were working on their open source implementation Stable Diffusion, but that took about a year until it became actually usable with v1.5
#aiart
Show threadLord Caramac the Clueless, KSCMar 14, 2024
@modean987 After SD 1.5 went public, all kinds of free image generator websites appeared, and I tried a lot of them, but my main platform is still #DeepDreamGenerator where I had been active since well before latent diffusion was a thing, when we only had #DeepDream and #DeepStyle. Then I discovered #Yodayo, a platform with over a hundred SD models now (mostly SD 1.5, a few SDXL), but almost all of them exclusively trained on anime, manga, and a few on Western comics and animation.
#aiart
Show threadLord Caramac the Clueless, KSCMar 14, 2024
@modean987 I've tried many other platforms, but there's only three I still use today: #DeepDreamGenerator, #Yodayo, and #Leonardo. Leonardo.ai has got only a small selection of SD 1.5, SD 2.1, and SDXL models, but it produces very nice results. Yodayo is uncensored, and there is a lot of erotic manga there, although many users publish only SFW images. DDG isn't very flexible and lacks a lot of features, but the user community is very nice, and you can combine SD with the older AI tools.
#aiart