“The fact that the faces all look sort of wrong doesn’t matter if half of the face is rotting into zombie flesh, and the fact that the bodies are constructed weird doesn’t matter if there are twisted extra limbs. Suddenly all of the flaws were working towards brilliant horror results, and I was getting some images back that didn’t look like anything I’d ever seen before — indeed, because no one had ever used the tech in this particular way. Start feeding those images back into the system and adding additional descriptions, and eventually you get some truly bizarre imagery. Working with text-to-image AI is less ‘creating’ in a traditional sense, and more like ‘mining.’ It’s a trial-and-error process of digging through infinity and finding the right pathways that guide the image towards what you want, while also taking many unexpected detours along the way when the AI has ideas of its own. For me, it was a new technology to ‘glitch,’ and horror turned out to be the perfect way to glitch it.”
Source: https://www.buzzfeed.com/daniellaemanuel/artist-ai-met-gala-body-horror
