📱💻 Join our July EdTech PD sessions, explore powerful tools, and earn up to 6 paid hours! Sign up now! -->

AI Prompting Guide

The information here is designed to help you get the results that you want out of AI generation. Bear in mind that AI tools are just that - tools - they are currently not often "perfect" or necessarily good enough for your specific needs. However, with some practice, AI can become a helpful timesaver that can streamline your workflow and free up valuable time before and after school when you have other things to worry about in your life!

Just like learning to type or using search tools, prompting takes a bit of effort to learn. Practice makes perfect! If you don't get the results you want right away, go back to your prompt and think about what might be leading the AI down the wrong path. Sometimes switching a word or two out will make a huge difference.

Back to AI Homepage

Different Kinds of Prompting

Prompting comes in two forms. Generally speaking they are interchangeable and can be used on any AI tool, but some tools respond better to one type over the other. Text generators often work better with what we'll call "natural speech" prompting and image generators often work better with "tag" prompting. It actually a bit more complicated than this, but for our purposes we'll keep it as simple as possible.

Natural Speech Prompting

Natural Speech Prompting is as it sounds - you write a prompt that sounds natural to how you would speak to another person. For example, if you wanted information about giraffes from Google Gemini, you might prompt something like:

Tell me some interesting facts about giraffes.

What are some things that might surprise me about giraffes?

List ten things about giraffes.

This sort of prompting is simple, effective with most text generators and will often get you the response you're looking for. Of course, just like with natural speech you can get much more complex:

In the style of an encyclopedia entry, give me a detailed description of giraffes and their morphology.

Write an outline for a children's story about a giraffe who is too short to reach trees and seeks help from other giraffes.

Create a Biology lesson plan for 10th grade students, at a 7th grade reading level, that covers the evolution of the giraffe's long neck.

Each of these prompts will result in different responses, and you'll often find that the more detailed your initial prompt is the more detailed the AI's response will be.

Tag Prompting

Tag Prompting attempts to utilize the "tags" that AI uses to analyze its own database. In simple terms, it's an attempt to "think like an AI" in order to get the most accurate response. This is more challenging to learn and requires more experimentation. For example you might prompt something like this:

masterpiece, oil painting, in the style of Monet, a rainbow-colored giraffe standing in a field, blue sky, cotton candy flowers, peaceful, relaxing

photograph, canon mark IV, f1.5, a giraffe eating leaves, acacia trees, African savannah, soft focus, distance blur, gorgeous, detailed

As you can see, this is not a very natural way of typing. However, this is easily recognized by AI and it is quite simple to tweak these prompts to develop better accuracy in your results. In addition to this, tag prompts also use "temperatures" or "weights" to help emphasize specific parts of a prompt. In general, the beginning of the prompt is naturally weighted as more important than later parts and numbers can be used to specify weights. These are on a 1.0 scale where 1.0 = 100% and .8 = 80% and 1.2 = 120%. It's best not to go too high or too low on these. This is what it might look like:

(photograph: 1.2), canon mark IV, f1.5, a giraffe eating leaves, (acacia trees: 0.8), African savannah, (soft focus: 0.6), (distance blur: 0.6), gorgeous, (detailed: 1.1)

Tag prompting also often includes "negative" prompts. These are prompts written to inform the AI what you DON'T want in the final product. Both "positive" and "negative" prompts are used to create very specific images.

Image generation can get far more complicated than text generation if you decide to pursue it.

Tips for Better Prompting

Provide Context

When prompting if you give some context you are liable to end up with better results. Asking for "information on giraffes" might get you very generic and surface level information. If you say something like "You are an experienced wildlife biologist specializing in large mammals found in Africa. Based on the impacts of climate change and human intervention what is the outlook of the giraffe species?" you will get a far more specific and robust response.

Be Specific

Try being more specific when prompting. Using our "information on giraffes" topic - saying something like "Give detailed information on giraffes" is less specific then "Based on research in 2022, what are the implications of climate change over the next decade on the population of wild giraffes in Africa." Directing the focus of the AI helps it provide richer responses.

Building On the Conversation

One of the coolest features of Text Generators is that they are often Chat Bots - what this means is that they are designed to "remember" what you have been prompting them with over the course of several prompts. So if I had asked something like "Give detailed information on giraffes" I can later ask in the same chat, something like "now provide a funny fact about them" the AI will recognize that "them" means giraffes and will provide relevant information.

You can often continue to build on previous prompts for a very high number of responses. Though there is a limit to how much an AI will "remember" over the course of a conversation. If it "forgets" something you can prompt its memory by saying something like "remember..." and copy/paste something from earlier. It will then remember that information in future prompting.

Be Aware of AI's Flaws

No matter how well you write prompts and how careful you are in building on them - AI is not perfect. In fact, there has been recent articles about how AI may never be able to fully avoid "hallucinations" or false information. The term for this is "slop" and AI can generate it within the first response or in the fifteenth response or even one sentence in one paragraph of a response. So it is always best to pay attention to what it is writing and verify from an outside source that you trust.

Image generators also struggle with things that many artists struggle with - eyes, hands, and poses. Prompting and re-prompting can help, but even the best image generators still create many flaws in the end results. The only way to work around this is to get deeper into AI generation prompting and even concepts/tools like LORAs, Image to Image prompting, ControlNets, OpenPose, and Generative Fill with Masking. You'll find that the best AI users have developed nearly as many skills as an actual artist does - many of them have art or programming backgrounds - so don't be put off if you can't get the best results.

Page updated

Report abuse