P: Generated images violate user guidelines

when I was doing a scene, I typed in child. VIOLATING GUIDELINES.    I typed in kid... Works fine.  Bloody daft !!

I am not even doing anything in the body of a person now. See the white area I am selecting. I Just need to extend the background of this photo (a classroom with children), so the image adjusts to the different formats I need to generate and it won't let me, telling me it violates the guidelines.... I definitely understand the need to regulate these tools, but the restrictions need to be smarter.

I'm sure 99.9 % of us have dozens of violation errors for no reason what so ever. Ignore them and carry on as normal

I was trying to extend the image below, and GF did fantastic on the left, right and top, but when it came to the bottom, I got a violations error!  THAT IS HER ARM!!!  There is NOTHING wrong with this image and it should not have been flagged. And thanks to AI's failure to know what an arm is, I have multiple violation errors reported on my account!  Remove them, and fix this.

So I'm aware generative fill is not meant to create nudity, but i have a bunch of sexy pictures where some of them include nudity and the images themselves are smaller than I'd like & i wish to extend the canvas to my preferred desktop size (2560x1440) & then have the generative fill, fill out the background.

Is this still possible and okay in regards to the terms of service?

I don't want to start a subscription only to find out I can't use it for the main purpose i wanted it for.

Adobe Photoshop beta is rejecting pretty much all generative fills that involve people's skin. It says I am violatiing Adobe's rules but I am not. These images do not contain nudity or any other undesirable attributes. Can you please help?

Generatives Füllen ohne Prompts (also ohne Texteingabe) erzeugt fast immer einen Richtlinien Verstoß und funktioniert nicht.

@CShubert 

I think the real problem is all those guidelines are not good for customers and sometimes they are discriminative.

chubby and fat for example violate the guidelines, a fat person is against guidelines? fat belly is a like a combo of guidelines violations...
In what world "Belly" is against some guideline?? we can do it manually with liquify etc, why not making it easier with ai?

Why all anatomic names are forbidden? belly? belly button? really? i can't add a belly button, but i have to hope generative Ai will produce one random using other prompts? and this is for almost all anatomic parts: breasts(not adult industry people don't have breasts?), torso, chest, abdomen... 

i think you should focus on generative results quality instead of banning all the words in the world.

About the quality and deformity problems, there are solutions (yes i would prefer better automatic generation too), you need precision and a little time but at least you can work on it.

1 Make sure selection (also working on entire backgrounds etc) is quite precise.

2 Make selection "bleed" some pixels inside the object around, for example, if you want to convert a long sleeve tshirt in a short one, select all the arm, make an expand selection over borders of few pixels (usually high res photos 15px is enough, low res even 5 pixels), and make sure that a little skin from the hand is selected, try with prompts like "skin" "short sleeve" etc.

3 Sometimes you have to work in pieces, cause AI doesn't accomplish in its randomness. Using the example above you have to divide the arm in pieces, starting from the hand, cause the AI often takes what it has around, so if you start from the hand, taking skin, and replicate for every piece, at the end you should have a complete arm with short sleeve, if you have some differences in skin, now that is all skin you can select the strange stains (don't forget to bleed pixels in good skin) and put "skin" or "." as prompts and try.

4 for some things you have to make a particular selection, for example an hand with a baseball bat, you have to contour the hand, give the selection a shape similar (but rounded) to that of a hand with a baseball bat, than add a selection of a shape similar to a baseball bat (it can also be a rectangle or elliptic, but it should contain a baseball bat), if it doesn't work at first shot, replicate steps above.

5 sometimes you have to work with pieces of copy and paste, for example you are working on an hand, 3 fingers are good, 2 deformed, take the 3 good ones, copy (copy also what is around, like space between etc) and paste on the original, merge levels, now you have an hand with 3 good fingers and the old one merged, select the rest of the old hand, make sure selection includes some pixels of the good part, and generate again.

think that most important problem now is guidelines

True enough. I often get the feeling that the people complaining most about Gen Fill have little or no prior experience with text to image AI. It's just something that's inherent in the majority of AI platforms.