Re:Stable Diffusion 2.0 and the Importance of Negative Promp...
by visarga on Tuesday November 29, 2022 @03:24AM
(#33785002)
Whatever content is avoided in AI training is going to end in the bin. It won't be referenced, the ideas won't propagate, it will get much less attention. I bet artists will start tracking how many times their names have been conjured up in prompts to demonstrate their impact.
Re:Stable Diffusion 2.0 and the Importance of Negative Promp...
by pmarreck on Thursday December 01, 2022 @10:41AM
(#33817719)
When I first tried inpainting, I used an image of this old poster for a game called "Myth II: Soulblighter" to see if I could expand it into a larger version.
The request was blocked because "violence was detected". It was a hand-drawn image of a video game boss attacking others with a scythe. I'm a 50 year old guy, I'm not a 10 year old boy, and I don't need to have "violence" (seriously? a hand drawn painting of a video game scene??) censored from me. This nanny-state upstream censorship is BS.
At least, if I ran SD 1.5 locally, I could render whatever I wanted to again, but now I can no longer do even that if I use the 2.0 model. This is dumb.
First they should sort out the legal question of training the AI on copyrighted material and propose use cases that the general public will find value in. Censorship can be dealt with later.