Skip to main content

GenAI Prompt Guide

Text- and image-based GenAI tools can help you solve problems, analyze data, and create innovative solutions. These tools work by receiving prompts to start or modify actions.

A prompt simply is a message or a command that you give to the GenAI tool to tell it what you want it to do. This page will help you get started using prompts for both text- and image-based GenAI tools.

Prompt help for text-based GenAI tools

Remember: The text-based GenAI tools that are most popular right now (such as ChatGPT and Google Gemini) are based on large language models (LLMs). These types of tools are not really subject matter experts. They're primarily great at language and content production and prediction.

For Stanford faculty and staff, you can try out what you learn with the AI Playground pilot.

AI Playground

Here are a few ways to use prompts with these types of tools:

Begin your prompt with a clear instruction or question


For example:

  • If you want the GenAI to write a short story, your prompt could start with "Write a short story about..."

Specify the desired output

  • Clearly define what type of final format or product you want the GenAI to generate.
  • Whether it's an outline, an email draft, a paragraph, an essay, or a dialogue, be specific in your prompt.
  • You can also experiment with specific word count parameters.


For example:

  • If you want a poem about nature, your prompt could be, "Generate a poem about the beauty of nature."
  • After describing what you want, try adding, “Present this in the form of…” and your preferred output.
  • For a word count parameter, try, "Write a paragraph with no more than 500 words about ..." or if you already have a result, you can use, "Shorten this to less than 500 words."

Clarify role, tone, and audience

  • Try giving the GenAI an assigned role and audience.
  • Experiment with assigning a specific tone or style.

For example:
  • “Act as if you’re an elementary school teacher talking to a class of fifth grade students."
  • “Use calming language.”
  • "Make this a little more friendly.”

Correct and give feedback

  • Try adjusting your prompts multiple times to see how it influences the generated text.
  • Give clear and straightforward feedback to hone the results.
  • Tell it what parts of the output were useful and what parts could be improved.
  • If you notice it got something wrong, note this in the thread, so it can correct any mistakes. And highlight what worked well.

Prompt help for image-based GenAI tools

Image prompts serve as the visual cues for the GenAI model to generate images based on your specifications.

A good rule of thumb is to include the subject, details, style, and format in your final prompt.

Take, for example, the below prompt progression, using Adobe Firefly:
 

GenAI prompt progression for image creation​

In this example, we can see the changes from a simple start, to adding a mood and setting, then more detail and style, and exporting in the desired format (an image).

Here are those steps at a high level:

Upload and/or describe the image you want to generate (or modify) in detail

  • Include key elements such as objects, colors, and scenes.
  • Be descriptive and specific: Use descriptive language to convey the mood, setting, and composition of the image you envision.

Explore style and editing

  • Generative AI can create images in various styles, like photographs, paintings, cartoons, illustrations, pencil drawings, and much more. Try a few style options.
  • Some image-based AI tools offer their own style options. Look for various options.

Iterate and refine

  • Generate multiple versions of the image by tweaking your prompts and experimenting with different settings.
  • Refine your prompts based on the results you receive, gradually improving the quality and accuracy of the generated images.

Experimenting, responsibly

Now that you have the hang of prompts, learn a few ways to begin experimenting.

And of course, always keep in mind our Responsible AI guidelines as you go.

Use Cases for Experimenting