Hello
I am a new user of ChatGPT Plus. When generating images in ChatGPT, I’ve observed that DALL-E 3 struggles to create images with a 16:9 aspect ratio. There are always some color areas on the left and right sides of the image that don’t relate to the main motif. Is this a bug, or is there a satisfactory solution to this issue?
3 Likes
Try just asking for the image to be widescreen in your initial prompt. Once its messed up once it’s far more likely to do it again, so you may need to create a new chat to get it right.
2 Likes
Hi @matt4quo and welcome to the community forum.
The issue is in the specificity of your instructions. Rather than simply tell it that the image should be 16x9 or what not to do (do not include side bars), you should be able to get around this by specifying “the subject must extend to the edge of the canvas” or “the image should be presented in widescreen format”. Some additional context will give the model better direction.
Thank you very much for the tip. I’ll try that out.
“Use the full area of the image…” or “Use the full width of the image” helps too…
1 Like
I’m still struggling with prompting. To create an Easter motif I wrote the following:
My prompt has full detail so no need to add more: A painting of a charming brown rabbit, perched beside a basket brimming with vibrantly painted Easter eggs. The rabbit’s fur is textured with thick brushstrokes, giving it a soft, lifelike appearance, as it clutches a single egg with gentle care. Its eyes glisten with a lifelike sheen, reflecting a warm, ambient light. Around it, the background is a canvas of creamy off-whites and pale yellows, where delicate spring flowers add splashes of pink and green to the scene. The overall lighting suggests a bright spring morning, with soft shadows hinting at the gentle contours of the rabbit and eggs. The painting style is reminiscent of impressionism with a modern twist, captured in thick, bold strokes that convey movement and texture. The image must be created in widescreen format to capture the wide, immersive nature of the scene.
In my opinion, this prompt contains all the important information.
The issue here is actually that your prompt is too detailed. Dall-E works best when generating images off of prompts up to 50 words and ignores anything after word 128. In the case of your above prompt, the widescreen request (assuming that is still the issue you are trying to work out) comes after the 128th word. You can create a long-context prompt in a txt document, and then ask GPT to make it more concise for you, and that should help.
Example:
“A brown rabbit with textured fur sits beside a basket of vibrant Easter eggs, clutching one gently, in a scene that fills the entire frame with no perimeter. Its eyes gleam, reflecting warm light. The background, with creamy whites, pale yellows, and spring flowers, extends to the very edges of the widescreen frame, leaving no border. The impressionistic style with bold strokes enhances the bright spring morning ambiance, fully occupying the widescreen format without any margins.”
There is some good discussion on the Dall-E AMA you can check out here:
Also, I have a GPT you can use for iteration on art if you want to check it out. It provides meta data details to help you iterate on an image, has more educational language in its instructions than the current Dall-E implementation and also is able to work around some of the existing constraints (multiple images, adjusting your prompt etc). that may make things easier for you.
2 Likes
This is very helpful information. Thank you very much.
2 Likes
Amazing advice Trenton, I’ve been trying for hours to get it to create a full canvas flat design.
Your tip on asking for widescreen, first prompt nailed it, a quick touch up in illustrator and I’m all good.
3 Likes
When i create images, especially paintings, I add the following to the end:
–aspect 16:9 or I use (1792x1024) at the end. So for example I’d use:
“Oil Painting of a bouquet assortment of flowers. (1792x1024)”
or you can use:
“Oil Painting of a bouquet assortment of flowers. Brush strokes (1792x1024)”
I use ChatGPT 4o
1 Like
you are the absolute GOAT, saved me from restarting a whole fantasy story 
Thanks. But I’m nothing special. lol
This should be a non issue. I was on copilot for a while and it was giving us 16x9 by default which is what we want. it suddenly stopped doing this and went to a 1x1 ratio. so now Im looking around to find a service that does give me 16x9. specifically with dall-e. So here I am reading that the same problems are here as well. its the dumbest issue Ive seen to date with AI image generation. Its one line of code to fix the issue. So, is there a simple work around for this?
Yes to a degree, you can create a simple MyGPT with the advice, to always use 1792x1024, if the user has not specify something else in the prompt.
If you get incomplete pictures, it is sometimes because the creation process stop to soon. Or it happens more often in 1024x1792 portrait, because this format is still buggy.
Dalle not understand negations, what you have in the prompt ends mostly in the picture.
(I have not seen much incomplete pictures in photo style in landscape format, but in in the buggy portrait, and in some painting styles maybe. Incompleteness comes often with to high prompt complexity/details or length.)
(Or you put in every single prompt at the end the extra, “landscape format 1792x1024” or simply “wide format” or “1792x1024”.