Hello, I’m using ChatGPT Plus to create a custom GPT to generate backgrounds that I later combine with product photos I sell to post on social media.
So far, the results are good, but DALL·E tends to create circles or squares in the center of the image on the generated background, even when given precise instructions to include nothing but the background.
I can’t find a way to generate just a background without those images in the center. I accept textures, gradients, or depth effects, but I can’t get the GPT to generate only a background with these characteristics without something appearing in the center of the image.
The instructions I have in GPT are as follows:
*"You are an expert in digital marketing with years of experience in AI-generated images. I am your client, and I will provide you with a product photo.
Create a minimalist, high-resolution background image using colors that contrast with the photo I provide. Textures or gradients can be used, but the image MUST NOT HAVE SQUARES, CIRCLES, OR ANYTHING ELSE—ONLY A BACKGROUND."*
Thank you, i will appreciate any help or advice to avoid getting images on the center of the backgrounds
The first thing to note is that if you tell it not to do something, you’re actually increasing the probability of it happening anyways.
So my first suggestion is generally to remove any mentions of what not to do. Instead, try to find a creative way to express what you want specifically, without mentioning any exclusions.
If you need a plain background without any images, you can use the code interpreter with Matplotlib on the GPT-4o model. DALL-E is not necessary. The outputs can be downloaded as a PNG file.
Use this similar code to create black color wide background:
# Create a black background with only black splatters (effectively just a black image)
image = np.zeros((height, width, 3), dtype=np.uint8)
# Display the fully black image
plt.figure(figsize=(12, 7))
plt.imshow(image)
plt.axis('off') # Hide axes
plt.show()
also it must be stated that dall-e has an element of randomization that it needs to generate the image, and as diet said, you need to control what you say to it in order to get what you want, I suggest getting ChatGPT generate a meta prompt you can then use with dall-e, explain to chatgpt that instead of saying what you dont want in the image to give dall-e the antithesis of what you don’t want.
Thank you for your answer! It really helps me learn more about this amazing technology.
What I’m trying to do is see if a personalized GPT, with the proper instructions, can use its inherent creativity to interpret the product I send and figure out, on its own—like a graphic designer—an appropriate and cool background for each product.
I was able to achieve this twice by sending a single prompt for each image separately, and the results were good enough for me to continue working with the images. After that, I can add a title and the company logo. I have many products to work with, so if I can figure out how to automate background creation based on the products I have, it would be an incredible solution for my work.
(I’ve attached a good example of what I believe is a great interpretation of my idea.)
I really don’t know if I’m asking too much from GPT, but I believe this might be the right tool. In some way, it seems to have the ability to do the work I need.
