Thanks I did this and it sucked - hence why I’m asking 
I’d like to understand / know any nuance in the actual prompt that was used.
In case you are still interested, the tip with the non-human monsters works. It is possible to create monstrous creatures without facial templates. However, with more aesthetic creatures, it is more difficult. It helps to describe the facial structure in detail, but even if it is very detailed, the silicon-like facial nose-mouthy template keeps showing up. I think I have to wait for an update in the hope that they will correct this.
Actually, the developers can trust their generator, it is capable of creating aesthetic and correctly looking creatures. They just have to let the system off the leash.
1 Like
Good to hear you found a workaround for most cases.
I maybe detected a other human thinking weakness. I created something like 10’000 images and always used trigger words like “photorealistic image” “hyperrealistic image” to avoid painting styles, and got always airbrush like images.
But now(!)… (i wish i would check this before!!!)
photo-realistic is mostly used form artist witch create a image as close as possible to realism, but it is still painted. If you create a photo image with a camera, you would not add “photo-realistic” to it. Instead you would have technical data in the meta data. I create mainly fantasy picture so the effect is not so unambiguous, and lens infos would probable reduce fantasy form the image(?). But try to use simply “photo image” or use even lens infos. I would guess you get more real live realistic looking images.
I would like to hear form people there experiences, with this little change.
1 Like
I’d read about another AI generator (Leonardo maybe)that it doesn’t recognize the negatives. So when you say " no sails," it just gets the sails. And when you repeat this command, you get more sails Maybe this is also true for Dalle-3?
1 Like
This is correct.
This is with the text shown passed directly to DALL-3. Were the ChatGPT AI trained or instructed better, it could rewrite a negation in the user’s language as an absolute omission of any mention in the prompt it sends, producing countering affirmative elements, but ChatGPT does not have that understanding.
1 Like
Great tips, Paul! I’ve also noticed that specifying the number of people or objects makes a big difference. Saying “two people sitting on a bench under a tree” gives much clearer results than leaving it vague. Also, adding details like “soft afternoon light” can really bring a scene to life. Looking forward to more tips!
I tried to collect some prompt experiences, focused on the weaknesses DallE still have, if you are interested. I try to find out how to overcome some issues. And it should give a beginner a kick-start, what otherwise will take long time to find out.