#DALL·E 3: Vertical Images are Wide Format Rotated Left

1 messages · Page 1 of 1 (latest)

fervent geyser
#

Bug Report:
Within ChatGPT with DALL·E 3, requesting vertical images of persons — especially full-length and especially women — results in wide format images rotated 90 degrees to the left. Images based on prompts with women are disproportionately more often affected by this than prompts with men.

Steps to reproduce:

  1. full-length shot of a stylish woman, outdoors, in instagram story format.
  2. happy woman enjoying her life, sitting in a café, photo in tall format.

Expected result:

Upright tall format images.

Actual result:

Wide format images rotated 90 degrees to the left.

Additional information

Browser: Firefox, Chrome, Edge ...
OS: Windows 10/11

vertical1.png vertical5.png
fervent geyser
#

full body shots of women are almost guaranteed to result in a rotated image.

vertical2.png vertical3.png
stray slate
#

+1 have the same issues whenever I try a tall/vertical/9:16 image creation

quick kelp
#

I had same issue. There are several open bugs for the same problem. Lets group them together.

surreal spindle
#

I am also having the same problem.
As a tentative measure, I first generate the image in square format.
Then, I fixed the seed value and prompt and generated the image again with the aspect "9:16".
It is time-consuming, but this method solves the problem with a high probability.

I hope this bug will be resolved soon.

fervent geyser
quick kelp
#

@surreal spindle Nice, however it is manual way to fix UI issue. This sort of feedback is useful for devs. Lest make sure we put votes at right places so they get to prioritize these bugzies. It is true that larger integrated model might make a lot of these things go away, and OpenAI might be there already next year, but it is also true that they are hitting ravenue streams and market cap that grant access to caputal to make all of this better at the same time. It is becoming a neat product already, yay

frail orbit
#

Mine aren't even rotated. ChatGPT claims it's vertical yet it just generates horizontally and cuts off the bottom and top. Every single time I tell it that it's still horizontal so change it to vertical, it then does the same thing while claiming the new images are vertical when they're still not.

lapis rampart
#

Dealing with the same issue, no matter how i adjust the prompt, no matter the modification request, it always comes up like that

frail orbit
#

Now the images are vertical.... except it rotates it, like the others mentioned and like the image examples above. So frustrating. I cannot use any image it generates because it's either square or horizontal and I need vertical. Will have to invest significant amounts of time to edit the images to blend the scenes naturally into a vertical image. Please fix this, thanks.

celest wing
#

Same here

IMG_8322.webp
lofty belfry
#

I have this same issue too. So much great art I can't use because it's not vertical

half depot
# frail orbit Mine aren't even rotated. ChatGPT claims it's vertical yet it just generates hor...

It's because ChatGPT (Dall-E 3) can't see or tell anything about the image generated, it can only read the text metadata Dall-E sends back to it... and Dall-E is (erroneously) reporting the image generated correctly, so ChatGPT confidently asserts this is the case because it trusts whatever Dall-E returned. This will all be a lot easier once we can do image-to-image and GPT-4V gets combined with ChatGPT (Dall-E). Obviously they need to also fix it now, but yeah.

gray spruce
#

Has been happening to me, I made a thread on this last week but didn't hear anything about it formally from the Team at openai

fervent geyser
#

Something's changed; with my initial prompt they turn out the right way most of the time now.

fixed1.PNG fixed2.PNG
glad igloo
fervent geyser
soft hull
#

@fervent geyser thanks it's very helpful