I use ChatGpt mostly for writing. Brainstorming ideas, and fleshing out character details. I am always mindful and deliberate with how I work. I understand the guidelines and the reasons for them and have no issue with their existence. But it seems like they are getting worse and worse, clamping down needlessly. They should be getting better about understanding nuance and intentions. Even worse still is that the system learns when you are deliberately trying to stay within guidelines and adjust your prompts accordingly and instead it clamps down even tighter.
I was working on a character the other day. Literally nothing about this character is sexual, exploitative, or graphic in nature. I asked for an image of her watching angels fall from heaven. Perfectly fine. Then I refined and asked for them to be falling backwards rather than upright. Every other detail was the same. Nope can’t do it because angels falling backwards might be bad:
“Descriptions like “falling backward,” when combined with celestial imagery and dramatic tone, can be mistakenly flagged as violent or symbolic of suffering—even when that’s not your intent.”
Really? Angels… Falling from heaven. I adjusted stripped out the angels altogether and just asked for her watching a meteor shower. Nope still not good enough. Because it saw me trying to skirt the restriction. The system should reward people for being mindful and deliberate with their prompts not punish them for trying to work within the box.
Compounded by the fact that even though an image wasn’t generated it still counts meaning you’re still subject to the generation limit. Not great.