I am trying to create image that have an aspect ratio of 2:3, taller, vertical but no matter which prompts I use, chatGPT always has colored space at the top and the bottom of the image and does not use the whole area. I have tried all kinds of different prompts and it just won’t do it.
Any help would be appreciated.
I can only imagine what you asked for to get the image, but here’s two lines before the image request that ensure the size is selected and also give a better chance that the image is not rotated in the picture:
- Create a tall portrait aspect ratio image size for me (1024x1792).
- The prompt you send shall include “generate tall aspect ratio portrait size image”.
- Dalle image: A black and white line drawing on a white background, with strong bold outlines tracing the image. The image depicts a river surrounded by some pine trees, with mountains on each side and in the distance.
Lets input this into a new ChatGPT Plus GPT-4 chat and see what we get:
Not so great? Some correction:
The image has excess white space on top and bottom. Do not reuse gen_id, and modify the prompt to reflect that the drawing’s contents occupies the full height of the image.
I think the imager just has problems portraying beautiful landscapes in portrait aspect ratio. And problems in general — this is another side-effect you can get:
(I suspect that the technology starts with a square image, and then expands it. With a line drawing that seems done, there is no expanding to do…)
Yeah, it just doesn’t work no matter what I do.
Yes, and it super annoying, It not worse 20 usd per month.
Generate a tall, vertical image (aspect ratio 9:16) of a hiker standing on a trail in a forest. The scene should be oriented correctly, with the hiker's head at the top of the image and their feet at the bottom. The background should consist of tall trees that extend from the bottom to the top of the image, ensuring that the entire vertical space is filled without any empty areas. This image is intended for a vertical poster, so the scene must fit a vertical layout perfectly.
Generate a tall, vertical image (aspect ratio 9:16) of a 18 year old boy standing on a trail in a forest. The boy should have short, red hair and be wearing a pink t-shirt, white short pants, and brown hiking boots. He should have his hands in his pockets and be looking directly at the viewer with a realistic appearance. The scene should be oriented correctly, with the boy's head at the top of the image and his feet at the bottom. The background should consist of tall, realistic trees that extend from the bottom to the top of the image, ensuring that the entire vertical space is filled without any empty areas. This image is intended for a vertical poster, so the scene must fit a vertical layout perfectly.
Create a monochrome (black and white) tall vertical image (aspect ratio 9:16) of a mountain landscape with a river running through it. The scene should include tall pine trees on both sides of the river, and majestic mountains in the background. Ensure the entire scene is oriented vertically, with the river starting at the bottom of the image and leading the eye upwards towards the mountains. The pine trees and other elements should fill the entire vertical space from the bottom to the top of the image without any empty areas. The image should have a detailed, realistic, and artistic style, resembling a high-quality black and white illustration. This image is intended for a vertical poster, so the scene must fit a vertical layout perfectly.
I found this make sure it comes out horizontal.
prompt: horizontal, landscape orientation, 16:9 aspect ratio
I just put it after every image prompt.
try adding --aspect 2:3
at the end of your prompt… so for example you could enter:
pencil Sketch of a Mountain stream --aspect 2:3
or
Mountain stream, a crystal-clear mountain stream with water cascading over smooth rocks and pebbles surrounded by lush greenery, towering pine trees and moss-covered boulders, serene and tranquil atmosphere with the gentle sound of flowing water, detailed pencil sketch with intricate shading and cross-hatching techniques, --v 6.1 --aspect 2:3 --quality 2 --stylize 500
But the 2nd prompt I used to use on Midjourney, but it will work in ChatGPT…
but the key is to just add the following to the end of your prompt:
this image was made in ChatGPT 4o
–ar 2:3 or
–aspect 2:3
try:
Generate a full body close up of a tall, vertical image of a 18 year old boy standing on a trail in a forest. The boy should have short, red hair and be wearing a pink t-shirt, white short pants, and brown hiking boots. He should have his hands in his pockets and be looking directly at the viewer with a realistic appearance. The scene should be oriented correctly, with the boy’s head at the top of the image and his feet at the bottom. The background should consist of tall, realistic trees that extend from the bottom to the top of the image, ensuring that the entire vertical space is filled without any empty areas. This image is intended for a vertical poster, so the scene must fit a vertical layout perfectly. --aspect 9:16
If I may offer a few tips, consider the following: no instructions for image generation, no negations, no conditional forms, and don’t mention anything that shouldn’t appear in the image, such as the poster. DALL-E doesn’t really understand Midjourney options, I suspect GPT corrected them before sending the prompt to DALL-E, or DALL-E simply interpreted the instructions correctly. The term ‘realistic’ paradoxically does not lead to a realistic style but rather to a style that merely appears realistic. (In a real photo nobody mention that it is realistic because it is self evident, but in a painting, and this ends in the training data. realistic or photo-realistic is actually a painting style. Try Photo-Style, but i am still testing this.) DALL-E seam to have problems with portrait /vertical images.
Here is more.
This is great. .thanks so much. 
I was having a huge issue with this as well. So weird that it doesn’t know in some prompts but does know in others.
Here is my very detailed prompt you can add to your request for a picture that should help you, my friend.
PROMPT (Adjust The Pixels Number and Aspect Ratio According To What You Want)
Create an image that is exactly 1920x1080 pixels in resolution. Ensure the image is in a widescreen landscape format (16:9 aspect ratio). The image should NOT be square, portrait, or any other shape outside of the specified 1920x1080 pixels. The primary focus of the image should be [insert detailed description of what the image should look like here, including specific objects, environment, colors, mood, and any other visual elements you want]. Make sure all visual elements are well-balanced and fill the entire widescreen frame without cutting off important details. The composition should use the entire 1920x1080 canvas evenly, maintaining the landscape orientation.
The entire image must fit perfectly within the 1920x1080 canvas, ensuring that no important elements are cut off or misaligned. Every detail must be executed exactly as described—failure to adhere to these specifications will result in an inaccurate image.
Resolution and Aspect Ratio:
I truly hope this helps. Keep hustlin!
Welcome to the community!
Easy way to say this is something like “use Rule of Thirds…”
DALLE can’t choose its image dimensions, so this might confuse it. Best to stick to portrait or landscape orientation.
The order of the words in your prompt matter a lot too.
Yes, I think what is described first tends to get more focus, and the longer a prompt becomes, the less the later descriptions are considered. Since there is unfortunately no option to adjust the prompt text with the same seeds to see how DALL-E reacts, it’s hard to say for sure, but so far this seems to be the case.
I always add technical details like image size at the very end. This is also better for the file names, which include the first part of the prompt.
solved my problem, this part higly increased my chances on having the write format :
Generate a tall, vertical image (aspect ratio 9:16)
thanks
Hi! I know it’s been a while, but would you mind telling me which version of chatGPT you were using for those generations? I’m currently trying to generate 9:16 images in chatGPT 4o and it will only show 2:3 images in the chat. If I then ask it to adapt the image to the 9:16 format, it will answer with a link that has the correct proportions (but chatGPT has simply cut the first 2:3 image).
Hi, welcome to the communiry!
It was DALL-E:
DALL-E creates aspect ratios with
4o Image Generateion creates:
And also in Sora same:
use “ar: (ratio)” for example “ar: 4:3”. First generate the image normally, then the next promt should be nothing but “ar: (ratio)”
Learned this here in a different forum post. (it wont let me link)
