Need a fail-proof prompt for full-length vertical portraits

[uneven sable]

Hello, I tried every possible keywords combination for 3 hours last night and it still fails 100% of the times which is unacceptable.

Definition:
• portrait means a person is represented,
• full-length means the person is visible from head to toes,
• vertical means the image orientation is vertical with the height being longer than the width (9:16)

Problem :
For some reason DALL-E 3 in ChatGPT does not represent portraits in full-length. I need it.
It also does not always orient the portrait vertically in a vertical image. I need it.
Those 2 errors combined means I get the correct output 0% of the time.
Below are samples.

What is the solution ?

[tulip basalt]

Unfortunately there's no fail proof method anyone has found

[uneven sable]

And what's the best method ?

[tulip basalt]

Best strategy I've found is to ensure the image matches the need for a tall image.. typically you want to make sure it's a full body shot by saying that or by describing the character as standing and perhaps describing their pants and shoes.

[serene forge]

Could you add all the keywords you used?
I have pretty good results by adding that it's a 'tall' picture.

[sudden remnant]

I am stuck in same problem too.🤣 tried every prompt that I can try with my brain and failed. desperatly hoping any solutions come out

[novel junco]

just call the figure/ full figured

full figured prompt

head shot/ bust image

[narrow cobalt]

I had decent results with "feet touching the ground". https://www.bing.com/images/create/a-full-body-portrait-of-a-blonde-farm-girl-from-fr/6541e6be66cb42b69160dafe9bd70115?id=ZTmKLVK2rSc5j4iBkf0ZyQ%3D%3D&view=detailv2&idpp=genimg&FORM=GCRIDP&mode=overlay

[sudden remnant]

square images are okay,but for somewhat reason dalle don't generate vertical full figure on tall image :/

[tulip basalt]

What have you tried so far?

[sudden remnant]

just tried some various prompts about full body figure,describing about floor or shoes, while requesting 1024x1792 sized tall image,it did give me full body images but rotated like this.

[tulip basalt]

Hmmm

it should work soetimes. If you click on the images in the desktop, you should be able to copy the exact prompt. If you can do that I can try to trouble shoot for you.

[sudden remnant]

nvm it's problem with current dalle
https://community.openai.com/t/orientation-problem-for-vertical-images/482759
https://community.openai.com/t/addressing-the-frequent-horizontal-orientation-of-vertical-images-in-dall-e-3/481903/2
https://www.reddit.com/r/ChatGPT/comments/176vjli/dalle_3_in_chatgpt_drawing_content_of_images/
https://www.reddit.com/r/OpenAI/comments/17ksui0/why_does_dall3_keep_generating_my_portrait/
https://www.reddit.com/r/OpenAI/comments/176vhlm/portrait_mode/

[hardy barn]

I have good results with adding " The full-body portrait... in a zoomed-out perspective". At least for the full body. Doesnt help for the rotation...

[dusk vector]

If resolution is Height x Width, does it not follow instruction if you swap these numbers? If it’s making the characters sideways, what happens when you tell it to use a resolution that is also sideways?

[hollow vine]

I am coming here to mention the same thing. Funny, I made x and check mark thing as well. Mine is also adding items to the photo, and there are clear instructions not to do so. The x and check mark helped when I was trying to correct the issue, but eventually, it just kept happening. I hope they create a workaround for this. I have been at it for the last couple of days.

[jagged moth]

"tall portrait-oriented image" this seems to help to some degree.

[uneven sable]

I came to the conclusion that the orientation is entirely based on the orientation of the reference images. eg if you ask for a tall image with a purple background and purple background images in the model are mostly wide, you'll get the wrong orientation.

[graceful oak]

Don't say Portrait or full length say "full body"
don't say vertical, say the specific aspect resolution in pixels

the terms portrait, full length and vertical can be miss interpreted because even though you are right these words can have slightly altered meanings

talking to GPT is almost like coding you have to say as few words as possible yet be as correct as possible

you would want to say
generate a full body image of a dog at 1080 x 1920.

however the largest aspect resolution GPT will output is 1792 x 1024

so you can ask it to generate at that aspect res then tell it to use the code interpreter to expand it to 1080 x 1920 if using gpt 4.

[hidden parrot]

This is the image I get with this prompt. Been happening all night after changing the prompt to everything I can think of, even asking ChatGPT 4 to generate the prompt for me after IT understands what I'm asking for. Please help?
"Please create a tall portrait-oriented image of Auril who is depicted as a lithe, attractive human woman with blue skin, a body made of ice and snow, and long white hair. She wears a thickly furred gown, surrounded by frost, and embodies arrogance and vanity, devoid of noble emotions like love or honor."

[tulip basalt]

Hmmm yeah interesting. I can't get it either....

[trail quest]

Unfortunately, I am also facing this problem. I tried every command to produce a 16:9 image, but it kept making mistakes and not taking the correct commands, and I don't know what the solution is. The bigger problem is that sometimes it also has a lot of trouble producing the details in these images, due to the high graphics in the background and not generating the person's nose style.

[urban surge]

I just use "aspect ratio 9:16" in my prompt ie "A photorealistic image of a winter princess, adorned in diamonds and platinum. Full length, 9:16 aspect ratio, quality:HD, style: natural "

[mellow jewel]

I have so many issues vertical images that basically most images are wrong do you have any new tips to improve it?

[sweet basin]

Whenever I try getting it to show a vertical picture it outright gives me an error message now

The very concept of verticality now seems to be crashing the AI

...I feel like I spent money for a bad product here, considering after the first week I had to fight hard just to get my images oriented right and now even trying to get them oriented right crashes the AI