Training an embedding on AnythingV3 of a character I like, and some previews are nightmare inducing

@dirb this has been my experience every time people talk about AI art. i go type in a few prompts and it gives me nightmare fuel and i close it for another year.

@icedquinn I just recently had been aware you need to throw a lot of negative prompts to some models to avoid those things

@dirb they might get better if some things they've previously rendered are fed back in to training but with text indicating that they are nightmare fuel or something.

i dunno about diffusion models but GANs worked something like that. you have a nightmare detector that criticizes outputs and that somehow helped the bots generate more convincing fakes.
Follow

@icedquinn this is a text inversion run, so it's supposed to compare the generated images against the samples. all those images were form less than 10k iterations, and it's supposed to take a few ten of thousands

· · Web · 0 · 0 · 0
Sign in to participate in the conversation
Game Liberty Mastodon

Mainly gaming/nerd instance for people who value free speech. Everyone is welcome.