that thread IMO is one of the top Dalle threads on this forum. And @PaulBellow ‘s page is a perfect complement to yours @Daller 
1 Like
IMPORTANT:
If you check prompts for triggers, ALWAYS use “don’t change the prompt, use it as it is.” in the prompt.
GPT changes the prompts often before they are sent to DallE. you must ask GPT to show you the JONS prompt sent to DallE.
1 Like
This is a drop code logic I use for that.
DallE JONS instructions
1: Never alter JONS prompt sent to DallE
2: Always show JONS prompt sent to DallE
@_j put me onto it.
1 Like
If you just want to see how GPT changes prompt this works
DallE JONS instructions
1: Always show JONS prompt sent to DallE
2: generate JONS prompt sent to Dalle in python frame after image generation
Only a thought.
1: Never alter the users entered text, use is as it is. (only translate it precisely to english)
2: Always show the user the JONS prompt sent to DallE
If the prompt is in JSON, it is already altered.
I learned, when I made my own GPT for this, that if it is not absolutely clear, it sometimes fails the task.
LLMs are not like linear programming languages, but it is still better to be as precise as possible, and order the advice correctly in sequence of process.
(And I painfully learned that GPT cannot handle nesting. It took me days to get this. I created a GPT for precise translation and no text alteration, but it still does not work 100%.)
Interesting that you get a Python, i get a JSON.
It always responds same way for me. I never trust it and I find I have to remind it sometimes, but on my end it seems to get it? Unless I put it in a core of a GPT then it don’t forget. My GPT redundantly check context in loops. . I build GPT I’m a novice though.
@Daller i see what you are saying. Yes that makes sense. TY
New logic.
DallE JONS instructions
1: Never alter user prompt sent to DallE
2: Always show JONS prompt sent to DallE
3: generate JONS prompt sent to Dalle in python frame after image generation
I don’t say its not working what you do, it is only from what i learned so fare… but i am not sure!
In a GPT advice list i would use:
1. Never alter the users entered text, use is as it is. (only translate it precisely to english if needed)
2. Prepare the JSON prompt for DallE with all needed field's and data.
3. Send the JSON prompt to DallE.
4. Always show the user the JONS prompt sent to DallE, after DallE returns the Picture.
Put this is all learning in progress… i could be wrong.
A other question, i have no access to Python, do you have a API account? I have only Plus.
For example, i can not convert the result in a other format. Others can. I don’t know why.
(i was writing when you write too… i see it now 
)
ChatGPT sends to dalle tool method with arguments in a JSON container. One of the fields is the prompt string.
By the way: stop saying JONS prompt! I don’t know who this Jon guy is, or why you are using his prompt.
This topic is not a “post a bunch of screenshots”. It expresses a concern over the content filtering of prompts, the cause of which being a particular prohibition on needlecraft patterns I displayed in screenshot #1, which I hope didn’t bring on this mess.
Another possible motivation for the prohibition on needlepoint is that bots run by bad actors could fill sites that sell patterns with derivative junk.
1 Like
I snorted lol 
Wife asked me who JONS is lol.
A other question, i have no access to Python, do you have a API account? I have only Plus…
Yes but I don’t mess with it. I’m teams and yes I do have access to python one of my GPT is a code writer…
Make sure you have code analysis on in the custom build emulator
1 Like
It is just a typo, i do 3 things at the same time. JSON.
I find it hard to imagine that needlepoint patterns are a billion-dollar business or even an existing problem. If anything, it would be counterfeit medications, but hardly needlework.
There is most likely no other reason for these erroneous blockings than laziness. Blocking based on simple trigger words makes no sense, and the lists were never revised or cleaned up. Having the names of well-known personalities on the list makes sense, but hardly ‘snow white.’ If the companies want to prevent Disney from losing a dollar, they need to analyze the context to determine whether it’s a Disney product. Color names and animal species cannot be trademarked and do not belong on the block list.
And having a system with no feedback make no sense not either.
And we have to explain now users here again, that they have to check the in-fact used prompt, because many not know that GPT alters the texts before sending them do DallE.
How many more articles about content policy are needed in this forum until someone finally improves the system?
1 Like
Code-Interpreter and Daten-Analysis
Thanks much! I completely forgot about it. I knew it, but I forgot that I had deactivated it. GPT could have told me…
1 Like
I ran into the same issue trying to make a 5k image of a cross stitch of myself playing music to animals. It was happy to generate a “needlepoint” of the exact same thing 
lulz I even googled if “cross stitch” is trademarked(it’s not).
OK “stop saying typos!”…
IMPORTANT!!! This is done with 7(!) words, actually 5 because 2 was “photo style”, non was violent, non was gore, non was usa related, non has the flag, non has anything political, No blood no flesh no weapons no war, non has anything abusive at all. And i have no time to waste to find out the tricks.
But THIS is your “security” system. 5 words + photo style.
“stop saying” it works. it don’t!
Switch of the string search blocking, for “snow white” and only use it for public peoples names. Give somebody a job to clean the list, you can not take all money in the grave.
How long we have to comment this in the forum. I will not waste anymore time on this. (Sorry @polepole, for this i am out. I only collect phrases and words, nothing else. You can post them, you know where, the links is on the page i worked the most.)
We are not baby’s, we should be able to make a horror pic for a Stephen King novel. but we don’t need more violence and perversion.
Take blood and gore out of the training data, for what it is there at the first place?
(In case, I blurred it, not Dalle. and i have NOT placed the flag!, i would put nightshade on this pic if i could.)
