TIED SHIRT

@dita see my review with "flat chest" image

@EDG Yay! I'll go check it out right now! Thanks EDG.

And yeah, the thing about AI is that it takes society's biases and preferences and reflects them back at us like looking in a mirror—after augmenting them and making them five times more visible. In that regard, it's a good I guess to see what our hidden biases are as a society.

I was reading an article just the other day that was about Elon Musk and his reign at Twitter, written by the person who used to be in charge of uncovering those biases in the company's algorithms—until Elon eliminated that department. He was talking about how Twitter used to have a feature to automatically select the part of a photo to use for a thumbnail when you upload it, but that many users were saying it was biased toward women and white people. And it's the same thing with latent diffusion models. The images the models are trained on are by and large of attractive, slim-to-average, white women—apparently those with large breasts, lol. So its knowledge of anything else is lacking.

That's one of the things that makes Stable Diffusion so much better than previous text-to-image AIs. Because it's open, people like yourself and others who post here can, with a textual inversion, emphasize a part of the data it already knows, or, with a LoRA, supplement the data with new stuff it didn't know before. (At least, that's how I've read TIs and LoRAs compared previously.) Eventually there will be LoRAs (or whatever comes after them… LoCons?) for everything, and people can choose their biases more freely, like we do every day outside of latent space.

@EDG A whale tail, when a thong sticks out of the pants