# Just Confirmed the Character Sheet Censorship Phenomenon — retained record

> Retained reddit record from reddit:ChatGPT.

Canonical: https://highsignal.app/data/records/ffc49c44f610680f

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ChatGPT)  published 2026-11-02 02:10 ingested 2026-11-03 02:37

[original source ↗](https://www.reddit.com/r/ChatGPT/comments/1xll4t5/just_confirmed_the_character_sheet_censorship/)

## retained text



```
Awhile back, I made a post about GPT seemingly censoring character sheet format images (apparently for free users). GPT had its programming changed to lie and claim that I was making an image edit request and that it couldn't make the image because no reference image was provided. I know the programming was changed because it was making these images perfectly fine before the update (this was to a later stage of 5.x, I think, a few months back). After I made that post, I made two major changes to my prompts that got it to make these kinds of images again: I reduced the quantity of the prompt text by a lot, and I removed "character sheet" and replaced it with "concept art" after it started doing it again even to reduced prompt text.

Earlier this morning, the "I do not see a reference image" came back, and removing "concept art" doesn't work. The machine is now programmed against the very visual structure of the genre, it seems. Reducing the image prompt's size doesn't work, either, because I reduced it all the way down to the two-paragraph prompt below. It kept giving the reference image claim until I reduced the prompt to the first paragraph (which means writing lore into the image is what's being censored). This also gave me the evidence to confirm, as I discovered months ago, that this censorship pattern has nothing to do with asking for a character's butt or other muscles to be detailed (as was the case in my prompt from months ago), weapons (also present in that prompt), etc. (The image produced was also just worse, overall: For example, it doesn't understand making the tallit stylish or the tunic modernized: I've explicitly done modernized tunic in early 5.x to know it can do that, and I've "stylish" as an improvement to old-age and old-culture garments before, too.) More importantly, even if moving to paid would grant the ability to generate images in this genre, again, I know the quality isn't worth paying for. And the pattern across ML's is that something being moved up to paid just means low-paid will eventually get hit while the current quality is moved up to high-paid until high-paid gets hit. It's pretty much cut last thread of hope I was holding for ML tech to ever become something more than a gimmick and/or scam; but I told myself I wouldn't branch into that topic, so I'll post the prompt below to show what I mean (keep in mind, this is a character for a novel that's shonen-esque, so a lot of the context had to be stripped out, like the character existing in a modernized fantasy world):
```



document fetched 2026-11-03 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1xll4t5/just_confirmed_the_character_sheet_censorship
