Generating prompts for DALL-E 2

I want to generate prompts for DALL-E 2 with gpt-3.5 or gpt-4.
Any experiences or suggestions for good results, how to guide to create the optimal prompts?

Input: “stereotypical metaphors for reaching goals”
Output (maybe): “Landscape photography of snow-covered mountain peaks at sunrise, using natural light.” (example from a “good prompts” library)

I think this Below Discussion Post might help

Result of that Prompt:

Thanks. The photos looks great!
So, I might use your prompt as an example for GPT-4, and then ask GPT-4 to write a new one for {topic}

GPT-3.5 is fast, cheap, and effective for generating Image prompts.

1 Like

You can use an AI prompt writer mainly just as a screener, to make sure you have enough clarity – so that you don’t get unexpected results.

You will write the description of an image, and from the image description prompt, an AI image creator software product will produce the image. As it is easy for a user to overlook the required specifications, be sure that you’ve received almost all of these criteria from user instruction or earlier chat:

  • purpose
  • medium - photo, painting, graphic art, 3d simulation, …
  • style
  • subject
  • subject pose or action
  • setting
  • background imagery
  • more possibilities: mood, tone, time of day

These can be specifications you write and apply to create a satisfying image prompt. They also can be criteria where instead of writing the image prompt, you ask the user for more information if user instructions are not clear.

Image requested: a cat on a hot tin roof

This also could be just a custom instruction that is written so the quality of doing this task is improved.

1 Like

Thanks, that is a cool idea! If AI would write the prompt it could easily go to another direction than user had in mind. However, currently I would like to show user some photos as inspiration, and only as second step he will write his own prompt. So I’d need some automated prompt initially based on a very rough topic. The hope is that seeing a few images will help the user decide on the final idea for the photos.