In my work related to visual and creatives areas, I sometimes like to appreciate errors. Same when generating images by AI .
I was in love when I realised that AI at first didn’t fully distinct man and women clothing, and used it randomly. In the first images of Mexican fighters I generated, many of them appeared with some fancy women clothing on, some even wore high heels. In case of the family photos, the funny thing was that some of parents’ faces were strangely similar to each other, and that the configuration was also quite random: 2 fathers, 3 mothers with one father (or uncle;) AI could add up to 3-4 wives, and basically came up with any combination that you can imagine. It would not judge you for willing to generate Jesus on plush pink cross, but forbid you from using the word “sucking”. Curious how, and in which way, with growing public use, the new rules and restrictions will block the freedom and “tolerance” of the algorithm.
Sadly in new versions of algorithm and it’s learning, this freedom is slowly fading due to “correct use” and lack of fantasy in users. As always the mass have rather “basic taste” and it’s very visible that the algorithm, obviously, is going in this direction.
From this particular reason especially, I like to use older versions of algorithms. I appreciate the unexpected. Mix of people, body features, and aesthetics. I like the unintended sense of humor that algorithm has in this area. What’s most interesting is how every version of the algorithm is different, and can create different outcomes.
trendhunting 