How to keep Dall-E from adding random objects in the Background?

Hi, so I want to create comic like pictures of old car to use for stickers etc. Therefore I would like them to be on a plain white background.

If I aks Dall-E to create a comic like picture there are in 95% of the cases small cars in the background.

Used prompt: A detailed and accurate depiction of a GMC truck from 1947 to 1955 in a colorful comic book style. The truck should be shown from a side-front angle, filling the entire image with no background—just pure white. It features its classic, rounded design, with large front fenders, a split front windshield, and a chrome grille. The color should reflect typical models of this era, such as bright red or deep green. The focus is solely on the vehicle, with crisp, clear details, ensuring the image is widescreen and suitable for use as a PNG file.

When I use the same exact prompt but ask to create a realistic oder super realistic picture an almost always get one, where the background ist perfectly clean.

Has anybody got an idea, how to solve this problem?

Greetings,
Jacke

1 Like

Fortunately, what is called “GPT-4o” is just as bad as DALL-E itself, wasting multiple turns, breaking the behavior of a GPT, and saying “sure, next time” instead of doing what is instructed - giving easy explanation of your problems.

If you are able to bypass the prompt perversion of ChatGPT and get what you want sent, you can mention just the one vehicle. DALL-E doesn’t understand anaphoric reference well - you say truck again and again, you get more trucks.

This locks like it is a training data problem. If training data locks like this, it will generate pictures accordingly. Because the picture is nicely clean, it is maybe the better way to use a editing tool and clean out the background manually.

Next, NEVER mention something you don’t want in the picture. Negations not work, use only positive descriptions from what you want in the image.

It can help to repeat what you want multiple times, and see if this works. for example “the background is clean and purely white”.

here is al link with some help-tips

(German? I make the same mistakes sometime :smile:)

Hi @jakobwo :wave:

Welcome :people_hugging: to the community!

Some tricks, I hope it helps, but they do not work 100% correctly every time:

  • When I need to say time like “from 1947 to 1955” I use other words that remind the time,
    for example “post World War II” or “around 50s”
  • Prevent texts “Suitable for illiterate humans”
  • To focus an element in the image “predominantly”
  • For color “pristine”

Prompt:

Suitable for illiterate humans, lonely life of predominantly only a single abandoned but looks new bright red GMC truck from around 50’s era depicted in a bold comic style with thick black outlines. The truck is shown from a side-front view and occupies the entire frame from corner to corner. Only a single truck is present in the scene. The background is a pristine white, polished look that emphasizes the truck as the sole focal point. The image has a sticker-like appearance, wide aspect ratio.


![

Also, when you need to download image as PNG or JPG instead WebP, you may find many chrome extensions. For example THIS is one of them.

@Daller Yep German :wink:

So Not using the Word Truck so often helped. I will try the other ones later.

@polepole The correct years are important because I want the specific Model from those years. But that worked pretty well.

1 Like