Using AI image generators to create graphics for your lessons

Posted by Greten on 12 Jun 2023 under Tips, Tools

There is a common saying that a picture is worth a thousand words. In education, this is true in many cases. Showing a learner what something looks like is better in learning rather than describing it using words. Teachers use pictures through posters and slide projectors, textbooks with illustrations, and elearning modules with digital photos, vector graphics, and even videos.

There are several ways to acquire images. Hire an illustrator or professional photographer if you want exclusive rights to your graphical elements. You can also buy them from stock collections. There are also free images on the internet, either those that already lapsed or are dedicated to the public domain, or you can use them in exchange for small non-monetary favor to the owner, such as giving them credit.

Artificial intelligence is a game changer with regard to acquiring images. There are several available AI-based image generators. All you need to do is enter a short description, commonly called a prompt, and click a button. The AI will then generate the images in the background and show the result to you. Some generators provide more than one image for a given prompt that you can choose from. Some have functions that let you do additional tweaking after generating the images.

A small robot holding a paint brush while painting an abstract art on a canvas.

For this entry, I will use and Bing Image Creator. Others might produce images of higher quality and come with more options. However, these two are free and easily accessible. You can use without creating an account or logging in, while you can use Bing if you have a Microsoft account which you might be using already in other applications.

Entering prompts in image generators

As mentioned, it's relatively easy to use AI image generators. All you need to do is to enter a prompt and press a button. While it is generally advised to put as many details as possible in a prompt to get closer to what you want, there's no guarantee that you will get exactly it.

Like several other AI-based technologies, image generators are trained using a large amount of data, in this case, a large number of sample images. If your prompt is too specific and cannot be fulfilled from the available data, it creates a "hole" in your image that the AI will do its best to fill out, creating an image that looks great but not exactly to your standards.

In Bing Image Creator, enter the prompt in the text field and click the Make button. Most of the time, it produces four images, although sometimes it produces less. For example, I entered the following prompt.

a nuclear power plant in a rural area

After clicking the Make button, Bing Image Creator provided me with four images below.

Four images of four different power plants surrounded by a grassy rural landscape in a 2 x 2 matrix.

I combined the images to make this entry simpler and more accessible. Bing actually created them as four separate JPEG files with a 1024 x 1024 resolution.

Doing a quick image search in both Bing and Google, I can immediately tell that these images show what a nuclear power plant looks like. Out of curiosity, I also tried using Google Image Search to see if the generated images would match an existing image online. Fortunately, there are none, making these unique images.

Using the same prompt in, it produced five images. However, I find only two of them useful in their current form.

Two images of two different power plants surrounded by a grassy rural wetlands in a 2 x 1 matrix.

These other three need some editing. In the first one, the reflected image on water does not match the reactor, but since it's near the bottom, we can just crop it. The second one has what appears to be fire or lava; either the AI confused the nuclear reactors to volcanoes due to their similarities in shape, or some of the data used were images of burning nuclear reactors. We can fix it using cloning tools in GIMP or Photoshop.

Three erroneous images of three different power plants in a rural environment in a 3 x 1 matrix.

The third one is tricky because the AI seems confused about whether to draw a larger reactor or a tornado. The error is so significant that editing it isn't worth the time. Feel free to discard such images. Anyway, the AI still produced other useful images.

That's how people work with AI image generators. Enter a description, and hope you'll get something close to what you want. Save those you find useful, edit if there are minor errors, and discard those with substantial errors. You also have the option to enter the same prompt again to generate another set of new images or tweak the prompt a little to eliminate mistakes and get closer to what you need.

Limitations of AI image generators

AI image generators produce a wide array of errors ranging from "this is useful but not what I expected" to "this has nothing to do with the prompt". The AI is limited by its data and tends to produce aesthetically pleasing images despite the error. This is similar to how large-scale language models (e.g., ChatGPT) will try to embezzle paragraphs with an articulate and convincing utterance that turn out to be incorrect if it has insufficient data to produce the correct statements.

The errors can be rare or frequent depending on what you are trying to produce.

Images with long and thin parts

AI image generators are unsuitable for creating images with long and thin parts, such as the human hand. The generated hands usually have an incorrect number of fingers and are longer than usual. If the hand is holding something, some fingers are too far and detached in a way impossible for humans. Other types of images prone to similar errors are helicopters and wind turbines.

Errors are frequent in trying to generate such images. If you are lucky, you will find at least one image with no glaring error, and it can still be useful even if it is different from what you're looking for. Often, you will find that you need to repeatedly enter the prompt and edit the image that you find useful.

Two erroneous images of wind turbines in a rural environment in a 2 x 1 matrix.

Using the images of wind turbines above, the one on the left looks like a mangled heap, as if a giant or Superman bent them like aluminum foil containers. The one on the right can still be saved. Only the frontmost wind turbine seems okay, while the rest are a twisted mess. On close inspection, however, the frontmost turbine is still not okay. Unlike the other two, the bottom left blade is not attached to the rotor.

Five wind turbines, each with three blades, in a rural environment.

Here's the fixed version of the image on the right. I moved the blade a little. Then, I deleted the other turbines and replaced them with duplicates of the fixed wind turbine.

The technology is progressing such that encountering this error is becoming rarer. However, it shows that AI cannot entirely replace humans. You're not just needed to enter a prompt but also decide which images to edit and which to discard.

Schematic diagrams

AI image generators, or at least the AI we have now, can generate beautiful stunning images. However, they cannot make schematic diagrams where specific parts must be in particular locations. For example, it can create an image of a nuclear power plant as it appears from the outside. However, ask it to generate a diagram of the inside of a nuclear reactor showing various parts such as uranium fuel and control rods, and it will fail spectacularly.

For example, the images below were created when I prompted BingAI to create a "block-and-tackle pulley".

Three erroneous images of block-and-tackle pulley in a 3 x 1 matrix.

A block-and-tackle pulley is a system with two pulleys and a common rope passing through them. However, all of these shows only one pulley, and the ropes are in the wrong places.

The workaround is to use the AI image generator to create the smallest and simplest parts of the diagram. In the example below, I prompted Bing to generate the rope, pulley, metal strips, and hook separately. Then, use GIMP to combine them into an accurate image of a block-and-tackle pulley.

On the right is a block-and-tackle pulley manually created by editing and rearranging four different AI-generated images: a pulley wheel, a rope, a steel strip with bolts, and a steel hook.

The legality of using AI-generated images

The use of AI-generated images is in legal limbo at the moment. There are several questions about who should own the copyright for the images. Is it the person who entered the prompt? Is it the person or organization who owned the AI? Or should they be considered co-authors? Adding to the complication is the copyright status of the images used by the AI as data. Should the authors or copyright owners of these images have the right to tell AI companies not to use their images to train the AI?

In the United States, the US Copyright Office (USCO) issued guidance saying AI-generated images are not copyrightable. That is, they are essentially public domain. It defines copyright as something that applies only to works of human creativity, and entering a prompt does not constitute creativity and is too far from AI-generated output. However, if a human author exerted a meaningful amount of creative effort into the final content, such as editing or modifying the AI-generated image, then the USCO will consider it copyrightable. The scope of the meaningful amount of creative effort has yet to be established. Hence, USCO intends to host discussions to solicit public opinions concerning copyright issues around the use of AI. For example, the images of the pulley I created and the wind farm I fixed will likely be entitled to copyright protection.

There's not much discussion going on about the copyright status of AI-generated images outside of the US. Still, many countries, including Germany and Spain, have provisions stating that only works created by humans are entitled to copyright protection.

In light of this, some AI image generators insist on using the images in specific ways, such as requiring attribution, for noncommercial purposes only, or can be used commercially only in premium accounts. How these policies will fair in court in the light of what USCO considers copyrightable is yet to be seen.

I'm not a lawyer, so don't take my word for it. Teachers and trainers using AI-generated images in their classrooms will likely be safe under Fair Use, just like in using any other image. There are two extremes you can consider in using AI-generated images. On one end is to respect the policies of all AI image generators you use when it comes to ownership and work your way around them. The other end is to lean on the statement of USCO that AI-generated images are not copyrightable, and thus, you can use them in any way you want. Note however, that as a creation without copyright, others can also use your AI-generated images without your permission. You may find a sweet spot between these two extremes.


AI image generators are valuable tools for creating educational visual aids. They are free and easily accessible, allowing users to generate images based on prompts or descriptions. However, there are limitations to consider. Overly specific prompts may yield unexpected results, even though the generated images are usually visually appealing. Creating images with long and thin parts or intricate diagrams can be challenging. Legal aspects are still being determined, with the US Copyright Office stating that AI-generated images are not copyrightable, and other countries may have similar provisions. Still, some AI image generators have their own usage policies. Educators should carefully tread these policies, existing laws, and the current lack of copyright protection when using AI-generated images.


Guadamuz, A. (2017) "Artificial intelligence and copyright", WIPO Magazine, retrieved 12 June 2023

Holt, K. (2023) "AI-generated images from text can't be copyrighted, US government rules", Engadget, retrieved 11 June 2023

Last updated on 13 Jun 2023.

Share your thoughts

* Required. Your email will never be displayed in public.

Instructional design and educational technology for effective learning