GIANT AI Guide for Educators: Simple Image Prompt Engineering Techniques
1- Use adjectives that describe your vision
Here are our favorite ones:
Emotion Words - Mood Meter developed by a group of researchers at Yale University has so many emotion words you can use. Bonus: you’ll enhance your own vocabulary as you try to communicate with AI tools. We, at The GIANT Room, LOVE the Mood Meter because it helps us to reflect on our emotions by gauging our moods and how much energy we have. For example, right now as I’m writing this article, I have a positive mood and somewhat low energy, so I would say I’m feeling calm. Two hours ago, I was in a very crowded subway that was not moving. I was definitely in a bad mood with very high, anxious energy. I was actually terrified to miss my next meeting. This morning, I went for a run and felt so positive and had high energy. I was excited to start the day. Finally last night, I heard some very bad news on TV. My mood soured, and my energy depleted. I was feeling sad and gloomy.
So, close your eyes and think of your cat! How does it feel as you’re imagining it? Add that emotion word to your text prompt as you instruct your AI tool to generate an image of your cat.
Descriptive words like color, size, shape, and structure. Here your art and math knowledge may come handy as you communicate with AI tools. There are more than 500 colors you can imagine your cat to be, and in many different patterns, shapes, sizes, or even geometric shapes! How about a blue cat with shiny eyes and an angular striped tail? If you have a clear vision of a cat in mind, try to use descriptive words to help your collaborator “see” your vision among the noise! Also, it’s sometimes fun to imagine “out of this world” creatures. So close your eyes, envision a creature of your dreams, and use words to describe your vision as best as you can!
2- Think of the “vibe”
Are you thinking your cat has a sci-fi futuristic vibe? Or more of a cute kawaii vibe? Sometimes a single “vibe” word has so many hidden meanings and AI tools may interpret it in different ways. But give it a try! After all, you need to work with AI to get to know it. Use any vibe you have in mind and see what the result may be. Here is a long list of “vibes” if you’d like to enrich your vocabulary in this department.
You can also try to be specific while describing the “vibe” you have in mind. For example, to you a “cool” cat may be wearing a dark pair of sunglasses standing on a skateboard. When you ask an AI tool to generate an image of a “cool cat,” it will draw its own “interpretation of cool” which may or may not be close to your vision.
3- Knowing a bit about Art History can help you A LOT
Again, one thing we surely know about AI tools is that they have been trained on more than 650 million images, so chances are they know about many, many, many different art styles and artists! You can think of it as if these tools have visited every single museum in the world and have studied the art pieces with much attention and detail. From cave paintings to ancient roman paintings, modernism, art nouveau, art deco, Bauhaus, pop art, and street art… you name it! If you ask AI tools to generate a cat image in any of these styles, they probably would know what you’re talking about. You can even name your favorite artist, and the tools might have an idea who that artist might be.
So, to communicate effectively with AI tools, it may be a good idea to start visiting museums, taking note of different art styles, and learning which styles interest you. Of course, you can also study art history and learn more about different artists and their styles.
4- If you are a creative STEM builder, you are a winner!
Another great way to “prompt” AI tools to generate images based on your vision, is to show them a sketch or a “prototype” as a reference point. For example, you may use an image of a 3D printed cat, a wooden laser cut cat, a cat made out of modeling clay, a cat made out of fuse beads, a cat with a box head, or a kawaii drawing of a cat as a reference along with a description of what you have in mind. Think of your STEM and maker skills as another superpower you can use to better articulate your vision for the AI tools.
5- Keep it short and to the point with words that compact lots of information!
When working with AI tools, you may use certain words that compact a lot of information without providing lengthy descriptions. In fact, some applications do not accept prompts longer than 400 characters (or even fewer) anyway. The good news is that a simple adjective like “aerial photography” includes information about the perspective, lens, and style of the image you have in mind. Below, you can see four images of a cat and prompts we used to generate them using Abode Firefly. See how changing just one word has resulted in very different images:
In the Classroom: How image-generating AI tools work
Monica Bernal from Robert E. Simon Elementary School in District 3 of Brooklyn introduced her 3rd grade bilingual classroom students to Adobe Firefly while teaching them about machine learning and AI. She gave her students an example of a prompt, “a cat drinking hot chocolate,” to illustrate how they could use the prompt tool to create their images. She then asked her students to create their own prompts focusing on animals. To her surprise, students came up with very unique images. In addition, some students wrote their prompts in Spanish and the software actually created the images they wanted. This was a great opportunity for those students that do not speak English.
Reflecting on her experience teaching AI to her students, she mentioned that next time she introduces the lesson to students she would do the following:
Adding a link to Adobe classroom in her Google classroom in order to keep track of her students’ work.
Providing students with opportunities to think about a topic that they are learning in class. For example, some students might be reading an interesting book and they could create their characters using Adobe firefly.
Showing students how to share, edit, and publish their creations.
Allocating at least 2-4 weeks for these lessons.
Giving students sentence starters to complete their prompts and create better designs.
Let students draw their designs first.
These are great tips and suggestions for other teachers who would like to introduce their class to generative image AI tools in a developmentally appropriate way.