AI ART Grousing
February 2024: "The irony is that I get less and less satisfaction as the AI Art programs get more and more powerful. It seems that they are all pushing more and more for the same archetype -- an anime version of Taylor Swift -- and away from my drawing style..."
It's like AI is pushing us harder and harder into bland GroupThink.
Last year I wondered if Midjourney was "training" on the images that we were generating -- which is known as "training on synthetic data."
December 13, 2023: "Does Midjourney "train" on the images I generate in their Discord? The next step of AI is to "train on synthetic" data, generated by AI...but will that mess up the results?"
Wouldn't that eventually lead to very inbreed results -- Habsburg AI? Sabine Hossenfelder thinks so:
- YouTube: Scientists warn of AI collapse
I played on this groupthink/inbreeding idea in August 2022: When I made an AI version of "America's Most Popular Painting," playing off Komar & Melamid's project -- "The Most Wanted Paintings on the Web."
That's why I think "training your own AI model" (like I did in Leonardo.AI, and later with the long poses) is the way to get ahead with AI ART generations. However perhaps I also need to write my own Art programs, like I did with GPT 3.5.
***
Sigh..was AI ART more exciting in 2022?
***
And now Claude 3 just came out from Anthropic, boasting that it it better than GPT 4.
No comments:
Post a Comment
Note: only a member of this blog may post a comment.