Uglification of characters

[bobhurl]

Have you noticed that the generation of images or videos tends to always make your female characters uglier (by making them more bloated, more corpulent) and older? I asked ChatGPT on the subject: it would not be wokism, but the result of the average of real images serving as models and feeding the base. I still have a little doubt. And I couldn't find an article examining the subject.

 

[MiG31]

Never noticed that myself on Grok, on the contrary the characters (male and female) tend to always be quite fit, there's not much variety in terms of body type if you're not specific in the prompt. Couple things I noticed on Grok though: if your prompt describes a woman in her 40s, the result tends to be women who look like they're in their late 50s. And if you specify a young man or woman in their 20s and use the Quality feature, the models tend to always be of a similar punk/hipster vein (think colored hair, piercings and tats) while they look a lot more neutral with the Speed feature

 

[Comfort Eagle]

Have you noticed that the generation of images or videos tends to always make your female characters uglier (by making them more bloated, more corpulent) and older? I asked ChatGPT on the subject: it would not be wokism, but the result of the average of real images serving as models and feeding the base. I still have a little doubt. And I couldn't find an article examining the subject.

You deserve nothing but disgusting gremlins for relying on AI image generation to begin with.

 

[Royaltickler2k]

Tbh my first thought was that Comfort was a bit harsh, but the fact that you were concerned about possible “wokism” in your AI-generated fetish material makes it difficult to argue against. Maybe “uglified” fetish material is some kind of karmic irony.

 

[sophilos]

I saw an odd promulgation when I specified the first of two characters as "Asian" and the second simply as "blonde". The blonde characters, as I generated more images from one prompt, took on an increasingly Asian aspect. Even when I added "Caucasian" to "blonde", both characters appeared Asian.

Okay, ethnicity and coloring are not exclusive categories. Still, I specifed mutually eclusive criteria only to get the same results. It was not until I deleted "Asian" that I regained control. And this did not happen when I specified "Black" for the first character alone.

Through enough iterations, though, the LLM will swap criteria between characters, add or distort figures and change the value of alternatives in drop-down lists. Its attention wanders.

I shouldn't wonder. When config.sys was a thing, we assigned the keyboard and mouse to INT6 and INT7. Even then, we humans were the slowest of external devices.