even with instruction in the GPTs :
The element will be presented in one pose and from a single point of view, without any text, typography, icons, or infographic elements.
prompt: advanced tank with railgun, anime cartoon style, blue as primary color
this tend to not happen in 1:1 square ratio
You may try this:
advanced tank with railgun, anime cartoon style, blue as primary color, no text, no typography, no icons, no infographics, single pose, single point of view --ar 9:16 --chaos 10 --quality 1 --style raw
Also you may visit to see about aspect ratio and text in images problems in following topics:
1- Bug with text in images in ChatGPT-4o
2- DALL-E 3 does not seem to understand “from behind”
3- Issue with Positional Accuracy in General Images
4- Aspect Ratio in Cha GPT does not work
5- GPT-4 with DALL-E image custom aspect ratio
6- Orientation problem for vertical images
Are such options understood from DallE? I found them only by midjourney.
Hi @Daller
I know that DALL-E doesn’t have parameters like MidJourney does. However, if it does use parameters and someone can explain how they work, I’d be happy to help understand it better.
Interestingly, I’ve noticed that when I use certain parameters, the results are noticeably different. I don’t fully understand what’s happening behind the scenes.
Here are the results from two sessions:
|Tanks much for the response!
Very interesting…
I try for weeks now to work around some issues.
The text nonsense sometimes drive me crazy. I even got images where the picture look like in a editing tool.
I have no experiences with MidJourney, but i will search a option list and check if it works in DallE.
I don’t want kidnap this post, but i try to collect some workarounds and prompt tips in one post, if you are interested. It would save me a lot of time to know all this from start.
(more words for the bot)
The way you made the result repeatable with each generation is a great help. I had never tried Midjourney-style prompting on DALL·E before, and those random artifacts were really driving me crazy these past months. Thank you for the quick and detailed response! 
(side question: What –chaos option mean ?)
DALL-E says (bottom of the image), but maybe hallucinates:
The --chaos parameter in the context of image generation controls the level of randomness and unpredictability in the output. A higher chaos value results in more varied and unexpected outcomes, while a lower chaos value produces more controlled and predictable results.
Here’s how it typically works:
Very, very interesting.
We’ve been asking for that feature for a while now…
Thanks for sharing!
WOW, it works!!!
I wished i would know this before, tanks much for the tip!
(I found 12 commands for Midjourney so far, will take a while until i check the useful ones working in dalle…)
here over 60 command option like (–xyz value) that Dalle claim to use
--lighting, --colors, --mood, --perspective, --fantasy,
--weather, --material, --movement
...
Not sure if all usable and at what degree of influence on the output, please share with us use cases (as the rate limit is too restrictive for excessive generation !)
DallE maybe “lies” / hallucinates.
Check a official list of midjourney like:
https://docs.midjourney.com/docs/parameter-list
So far:
Tested with photo-realistic image, but a fantasy scene.
Could be the style makes a difference…
(I created a reminder not to stop me with warnings. But GPT and DallE not seam to care much about advice’s in the reminder.)
I am testing right now…
I actually have a suspicion and a question. GPT changes the prompts before they reach DallE, could it be that it actually not understand the parameters, but know what they would do in midjourney, and so changes the text accordingly? Has DallE really get the “–chaos 100” or hast GPT change it with a suitable text?
We should use “don’t change the prompt and use it as is”. (i not write in English +“, only translate it to English.”)
Even if DallE not understands the parameters directly, it seam to give GPT a advice how to optimize the prompt. (?)
What seem works:
–chaos 100 (more intensive)
–weird 3000 (maybe more creative… but DallE complains about it)
–style raw (seam to have no effect, but maybe it stops the nonsense text. That would be great, the nonsense texts GOES ON MY NERVES!)
Seem not to work:
–seed 1 (no effekt?)
–quality 0 (no effekt?)
–stop 10 (but DallE complains about it, but no effect?)
I speculate now you get results with simple words witch work like options: “darker” “very dark” “chaotic” “dreamy mood” “be creative” “very insane 
”.
Just never mention something you don’t want, DallE don’t understand negations. “–no yyy” not work.
And if GPT modifies your prompts efficiently with midjourney options, use them.
But i think if you switch off GPT modifications, the options not work, or DallE somehow tries to interpret them same as “very chaotic”. “raw style” seams to suppress this stupid nonsense text…