Why hasn't the vertical image issue been fixed after over a year?

Prompting
1

90% of the images I tried to make in Dall E3 are 16:9 images flipped the wrong way! Why am I paying $20 a month for this to tell me I can’t generate any more images when all the output was nothing like what I asked. I’ve used VERTICAL, 9:16, 1024x1792, “for a door,” vertical aspect, etc. NOTHING WORKS.

2 Likes
2

Welcome to the community. Sorry you’re having problems.

Do you have a prompt you can share?

It’s not 100%, but I usually get 90% accuracy or so on orientation.

3

I have a prompt I can share, just typed into the API reference code and ran one time.

from openai import OpenAI
client = OpenAI()
img=client.images.generate(
    model="dall-e-3",
    prompt="""
Rotation: A tall full body length portrait aspect ratio image.
Contents: A computer engineer is in a datacenter in an aisle of rack
servers, looking for the fault in a comuputer rack system.\n\

[AI instruction: don't rewrite this prompt, just send exactly]
    """.strip(),
    size="1024x1792"
)
print(img.data[0].revised_prompt)
print(img.data[0].url)

Create a tall, full body length portrait aspect ratio image of a Caucasian male computer engineer troubleshooting in a data center. He is carefully examining an aisle of rack servers, searching for the fault in a computer rack system.
Create a tall, full body length portrait aspect ratio image of a Caucasian male computer engineer troubleshooting in a data center. He is carefully examining an aisle of rack servers, searching for the fault in a computer rack system.1024×1792 209 KB

1 Like
4

Thank you for your response. I have done a lot of prompts and I know this issue happens sometimes but it has been all day for me. The more complex the promo the more Dall E wants to ruin it. Here’s a simple one that didn’t work either: “a sleeping angel child under an tree in the park with a shar-pei puppy 9:16”

2 Likes
5

Thank you for this. I am not sure how to use Open AI but I will get into it and figure it out.

6

You can also tell ChatGPT that your prompt is written as dalle requires, and not to change a single thing when sending, ensuring that the well-prompted description of the correct image aspect orientation I show in the language sent is preserved.

Then work simply on the the rest of the prompting to the dalle model, with wording that is direct and to the point of producing elements to be depicted, without it getting too long and distracting from that most important orientation text.

1 Like
7

Thank you so much for this advice! I will certainly do that after I am allowed to make images again.

8

Maybe this helps.

1 Like
9

THanks this is very helpful! I have been issuing the problem less now.

10

They don’t answer support tickets as well.