Hi Everyone, first post here so please be gentle and apologies in advance if I am very noobish or get something wrong. Had a search around and saw a few posts about content policy conflicts and scary fictional content but I tried something and it makes me feel this content policy makes no sense. I’m a big fantasy fan and aspiring writer, trying to generate some dark fantasy content and I suddenly now get the “I’m unable to create that specific image due to a conflict with our content policy. However, I can help with alternative ideas or modifications to the concept. Let me know how you’d like to proceed!” message when trying to generate fantasy images. Here’s why this makes no sense to me. The prompt I used was very simple: “A white orc, riding a giant white wolf, holding a mace made of bones. On his back are spikes with skulls”. A bit scary sure, but not particularly horrific or gory - this exact character was in the Hobbit which is a PG12 film. I tried each component of the prompt individually - I tried “A white orc”, a “white wolf” (interestingly my first attempt at white wolf was refused), “A mace made of bones” and “spikes with skulls” - the AI generated all of these images individually without refusal. But when I put them together in one prompt to have them in one image, it refuses. This makes no sense to me on a technical level…but more than this, how are users supposed to create any kind of fantasy or horror/thriller genre content with such an extremely restrictive content policy? Are we only allowed to generate images of puppy dogs and rainbows and coffee cups in the rain?
Here is the full sized image if you wish to use it, you have my written permission to use this image in any way you see fit.
You could also upload this image as a reference and ask for more like it if you get more problems.
@jcgoris, would you be willing to share one of your prompts that was rejected? We can then examine why @Foxalabs was able to produce the image using DALL-E 3 while your request was blocked by the content filter.
Welcome to the club! The security system is absolute BS and makes no sense at all! The developers are not fixing it.
In my case, it blocked content that wasn’t even scary whatsoever. This part of the security system uses a stupid (no GPT skills) word block list, and any name ever used is blocked. So, names like “Snow White,” “Black Panther,” or “Nirvana” were blocked, and neither GPT nor the user gets any feedback on what triggered the block. The system doesn’t just block some dark images, but even 100% acceptable prompts.
The advice I can give you is always to check the actual prompt used, as GPT changes the text and sometimes inserts trigger words that were not in your original prompt. Then, search or ask GPT if there is any name ever used by any company.
Like you, I mainly create fantasy images, and all the dark or horror pictures I ever generated were things that could be used for a book cover, no blood, violence, or gore. But as I said, it even blocks 100% acceptable prompts. You had “white” in your prompt, so GPT might have inserted “Snow White” into the text for DALL-E, GPT did this in one of the cases in my input.
(In one case, a frog like creature licking his own face with a long tongue, was too much pfui…)
Sometimes DallE generates a picture, but simply not create the image like described in the prompt.
Hi @jcgoris 
Welcome to the community!
Your prompt works well. If you used a different prompt, can you share it, please?
Here was my exact prompt. See screenshots of where it was rejected. I tried each part individually and then it worked…so it doesn’t make much sense.
“A white orc, riding a giant white wolf, holding a mace made of bones. On his back are spikes with skulls” This was my prompt - it was rejected multiple times even though the individual components in the prompt were tried individually and none triggered the content violation. See screenshot below
Are you using a custom GPT to generate the images, because the ChatGPT logo has been replaced with a custom logo.
If yes, can you try to generate the image using vanilla ChatGPT?
Yeah, I see that too @vb; but it looks like @polepole is using the same GPT. I wonder, are you using the free version of ChatGPT? It might a have stricter limitations.
Here’s what I made with my dndGPT:
Try it on DndGPT, mate, I’m curious: ChatGPT
I have ChatGPT Plus subscription, paid version.
And I used DALL-E on ChatGPT.
Me too, the Logo is the DallE Logo, not the ChatGPT.
Hey there, am using Dall-E, by Chat GPT - nothing custom as far as I know. Are the ChatGPT and Dall-E content policies very different?
K. Well. I am 
biased, but dndGPT is the clear winner. 
It is important not only to analyze the entered prompt, but also to consider what the GPT system generates from it, before sending it to DALL-E. GPT modifies the input text, and it can happen that GPT itself adds blocked words to the prompts.
You can use "use entered text as it is, and send it unchanged to DallE" to avoid more confusion. because GPT changes the text, sometimes the same input triggers and sometimes not.
(forgive me my frustration, i spend too much time because if this trigger system. All this trigger system confusion is wasting time.)
The logo is from DALL-E. I use the Plus version and the browser to generate images (not API).
(I also use a self-made GPT with some instructions to balance out some weaknesses of GPT. But it was not possible to teach GPT do avoid or detect trigger-words.)
@polepole @vb @Foxalabs @Daller @thinktank See screenshot, my prompt is still being blocked.
The image @polepole generated is even more macabre than I was going for (but nice!) witht he landscape. I really don’t get it, I can generate images of skulls, orcs, maces made from bone, spikes etc. all individually but for some reason they’re not allowed to co-exist in the same image
