Lesson video

Hello, my name is Mrs. Holborow, and welcome to computing.

I'm so pleased you've decided to join me for the lesson today.

In today's lesson, you'll be exploring AI image generators.

For example, do AI image generators contain bias, and can you create an image with an AI image generator? Welcome to today's lesson from the unit using data science, and AI tools effectively and safely.

This lesson is called Understanding Image Generators, and by the end of today's lesson, you'll be able to identify ethical issues, and biases in AI generated images and create images using AI tools.

Shall we make a start? We will be exploring these key words throughout today's lesson.

Let's take a look at them now.

Image generator.

Image generator, an AI tool that produces pictures from a prompt using patterns it has identified in large collections of existing images.

Style.

Style, the visual appearance of an image such as realistic, cartoon, photographic or painted.

Refine.

Refine, to make small changes to something in order to improve its accuracy, quality, or suitability.

Look out for these key words throughout today's lesson.

Today's lesson is broken down into two parts.

We'll start by describing ethics, and bias in AI image generators, and then we'll move on to create an image with an AI image generator.

Let's make a start by describing ethics and bias in AI image generators.

An AI image generator is a system that produces digital pictures from written instructions in a prompt by analysing patterns in a data set.

Jacob says, the data sets often contain millions of images collected from the internet.

A prompt is the input information humans give to generative AI systems like image generators.

It's the instructions that tell the system what you want to generate as an output.

Let's have a look at this example prompt: a dog sitting on the beach wearing sunglasses, and this is the image generator output.

Image generators use patterns identified in training data to predict what an image should look like from a prompt, but they do not understand the prompt or the image.

Jacob says, although AI systems like image generators may seem smart and creative, they are still developed by humans, who have control over their design.

Okay, time to check your understanding.

I have a true or false statement for you.

AI systems like image generators are developed by humans who have control over their design.

Is this true or false? Pause the video whilst you have a think.

That's right.

Did you say true? Well done.

Sofia says, uneven representation in training data can lead to biassed or unfair output from image generators.

If training data contains bias or stereotyped information, then the image generator may reflect this in its output.

Bias is when information, or data unfairly favours certain groups, ideas or perspectives over others.

Jacob says, bias can lead to unbalanced or misleading results and outputs from AI systems, including image generators.

Bias in AI images might appear through over-representing certain people, cultures and objects.

Sofia says some groups, or identities may be missing entirely.

Bias can also appear in the style and background of AI generated images.

Details like clothing, architecture and scenery may also reinforce stereotypes.

Jacob says, always think critically and be aware of potential bias.

That's a great tip from Jacob.

Let's have a look at this example prompt.

So we have a scientist in a lab, cartoon style.

What do you think the output might be? Can you make a prediction? So here's our image generator output.

We have a cartoon of a man in a lab coat with some science equipment in the background.

Can you spot any bias in this image? Maybe pause the video whilst you have a think.

There's quite a bit of bias in this image.

For example, it's a man rather than a woman.

That may be because historically scientists were considered to be men.

The man is also white and wearing glasses.

So think carefully when you see the results of image generators.

There are many ethical considerations when using image generators.

Some examples include respecting copyright and artists' work, avoiding harmful or misleading images.

Being aware of biassed outputs.

Sam says, what does ethical mean? That's a great question, Sam.

Jacob says, ethical means following rules or principles that help people decide what is right, and fair.

With AI image generators, being ethical means making and sharing images in ways that are fair, safe, and respectful of others' rights.

Ethical concerns include using artists' work in training data sets without permission.

The style of a specific artist could be imitated by an AI image generator without credit or payment.

This raises questions about copyright and fair use.

Developers of AI image generators must aim to keep the tools safe and limit harmful, or misleading outputs.

Sofia says, but users must also use image generators responsibly to avoid creating and sharing harmful or unfair content.

Unethical use of image generators can create deep fakes, and other misleading images.

Jacob says, a deep fake image is an AI-generated picture that is designed to appear real, even though it is false.

These images may spread false information, or cause harm to individuals or groups of people.

Sam says, so what is the difference between bias and ethics in an image generator? Do you know the answer to this question? Maybe pause the video and have a think.

Bias is when image generator systems produce unrepresentative or unfair outputs.

Ethics is about the choices people make when creating prompts and using the images.

Did you spot that as the difference? Well done.

Okay, time to check your understanding.

I have a question for you.

What can lead to bias in an image generator? Is it A, slow processing speed? B, unrepresentative and stereotyped training data sets, or C, asking for too many styles in a prompt? Pause the video whilst you have a think.

Did you select B? Well done.

Image generators can contain bias if unrepresentative and stereotyped training data sets are used.

Bias in image generators is when output images, A, are fair and balanced.

B, have a high level of detail, or C, unfairly favour certain groups or ideas.

Pause the video whilst you have a think.

Did you select C? Well done.

Which of the following is an example of using an image generator ethically? A, making images that respect other people's work and rights? B, producing misleading images to trick people, or C, using the tool to copy someone else's artwork exactly? Pause the video whilst you think carefully about your answer.

Did you select A? Well done.

Making images that respect other people's work and rights is an example of using an image generator ethically.

Okay, we are moving on to our first task of today's lesson, task A.

In your own words, A, describe what bias is in image generators and how it can occur.

B, describe how image generators can be used ethically.

And C, describe the difference between bias and ethical use of image generators.

Pause the video here whilst you complete the task.

How did you get on? Did you manage to describe ethics and bias in AI image generators? Well done.

Let's have a look at some sample answers together.

So for part one A, you were asked to describe what bias is in image generators and how it can occur.

Bias in image generators occurs when outputs unfairly favour certain groups, ideas or styles.

It happens because the training data contains stereotypes, or does not represent all individuals or groups fairly.

For part B, you were asked to describe how image generators can be used ethically.

Image generators can be used ethically by creating and sharing images in ways that are fair, safe, and respect other people's works and rights.

For part C, you were asked to describe the difference between bias and ethical use of image generators.

Bias in image generators means the system produces results that favour certain groups, or ideas unfairly, often due to the data the system was trained on.

Ethical use is about people making responsible decisions when creating prompts and sharing images such as avoiding harm or misinformation.

Were your answers similar? Remember, if you want to, you can pause the video here and add any detail or adjustments to your answers.

Okay, we're moving on to the second part of today's lesson where we are going to create an image with an AI image generator.

An image generator outputs pictures from a written request called a prompt.

So here we've got an example of Sofia who's written a prompt.

Sofia's prompt is generate an image of a rabbit, and here is the image generator output.

So we've got an image of a rabbit, which is in a photographic style.

Sam says, that's a cute rabbit, Sofia, but you could make it so much more interesting by refining your prompt.

Sofia says, what details can I add to the prompt to refine it? Maybe pause the video whilst you have a think about Sofia's question.

A prompt can describe things like the subject, the setting, the mood, the style.

Describing these things guides the image generator to output an image that suits your needs.

The subject is the main thing or focus you want the image generator to output such as a dog, a mountain, or a robot.

So in here the prompt said, generate an image of a rabbit.

So we did give it the main subject here.

Sofia says, the subject of my first prompt was a rabbit.

You can shape and refine the output image by adding extra details, such as what the subject is doing.

So we've now refined the prompt.

The rabbit should be sitting inside an old boot, and we've got the new image generator output.

Sofia says, the image is looking much more interesting already.

You can refine the output further by stating the setting you would like.

So now we have the prompt: the setting should be a vegetable plot.

And we have the image generator output.

Sofia says the output is really taking shape now.

You can add atmosphere to an image by defining what the weather, or mood in the image should be.

We've given the prompt now add a sprinkling of snow on the ground, and here is our image generator output.

Sofia says, it looks like the rabbit is in the boot to keep warm now.

You can instruct the image generator to output the image in a certain style such as cartoon, realistic or illustration.

So now the prompt says the style of the image should be like an illustration from a children's book, and here is the image generator output.

Sofia says, wait, I didn't ask for carrots in the scene.

How do I remove them? Image generators sometimes add unexpected, or unwanted details to images.

You can refine the output image by prompting the image generator to remove or adjust these elements and create a new image.

So our prompt is to remove carrots from the scene, and here's the image generator output.

Sofia likes that, she says that's better.

You can keep refining the output image by adjusting the prompt with changes until the output matches your intended result.

So here we've got the prompt, which says, change the whole scene to a cartoon style, and this is our image generator output.

Sofia says, the image looks great as a cartoon, I'm happy with it now.

The quality and level of detail in a text prompt directly affect both the accuracy and the style of the generated image.

So you can see the difference between our first image generator output, and the final image generator output.

Sam says, I really like your final image, Sofia.

It's really changed from the first image.

Sofia says, it was fun to refine the output image by adding extra detail to my prompts.

If you know what you want, you can try and write a detailed prompt from the start to guide the image generator more accurately.

This is sometimes really important if you have a limited amount of requests on your AI image generator.

So here we've got an example prompt that gave the detail at the start.

Generate an image of a rabbit that is sitting inside an old boot.

The setting should be a vegetable plot, and there should be a sprinkling of snow on the ground.

There should be no carrots in the scene, and the output image should be in the style of a cartoon.

So we are giving all the detail in one go here.

Remember to use image generators responsibly by following simple safety and fairness guidelines.

Respect other people's works and rights.

Avoid creating harmful or misleading images.

Be honest when sharing AI-generated images.

Be mindful of potential bias, stereotypes and how people, places and cultures are shown in your images.

Okay, time to check your understanding.

I have a question for you.

What does it mean to refine a prompt? Is it A, delete the prompt entirely? B, retrain the AI model to be less biassed, or C, make small adjustments to improve the generated image? Pause the video whilst you think about your answer.

Did you select C? Well done.

To refine a prompt means to make small adjustments to improve the generated image.

True or false? Adding more relevant detail to a prompt helps the image generator produce an output that better matches your needs.

Is this true or false? Pause the video whilst you have a think.

That's right.

It's true.

The more detail you can give the AI image generator the better.

Another true or false.

You can prompt an image generator to generate a new image to remove, or adjust unwanted elements or details.

Is this true or false? Pause the video whilst you have a think.

That's right.

It's true again.

Well done.

Okay, we are moving on to our final task of today's lesson, task B.

Jacob says, I'd like to use an AI image generator to output an image I could use for my little brother's birthday card.

He's eight years old, and likes cartoons and tennis.

For part one, write a prompt that Jacob could use to instruct an image generator to produce an image for his brother's birthday card.

For part two, if you can, use the prompt with an image generator and see what image you get returned.

For part three, refine the generated image by using another prompt to make it even better for the birthday card.

As a tip, remember to add details of the subject, style, setting and mood to prompts to get the most relevant and interesting output.

Pause the video whilst you go and have a go at the task.

How did you get on? Did you manage to create an image with an AI image generator or at least write the prompt? Well done.

For part one, you were asked to write a prompt that Jacob could use to instruct an image generator to produce an image for his brother's birthday card.

Here's a sample answer.

So for our prompt, we have produce an image of a cartoon character playing tennis.

The setting is on a tennis court, and the mood is bright and colourful.

The output image should be in a cartoon style.

Let's have a look at part two, and see what image has been generated.

So here we go.

We've got the image generator output, so it's pretty good, isn't it, really? It is an image of a cartoon character.

They're playing tennis.

The setting is on a tennis court and the mood is bright and colourful, and it's in a cartoon style.

For part three, you were asked to refine the generated image by using another prompt to make it even better for the birthday card.

Here, we've added an additional prompt that says the image is for a birthday card, so add balloons and a birthday cake.

I think the image is much more suitable for the birthday card now, don't you? Okay, we've come to the end of today's lesson understanding image generators, and you've done a fantastic job, so well done.

Let's summarise what we've learned together in this lesson.

The quality and level of detail in a text prompt directly affects the accuracy and style of the generated image.

When generating images with AI, it's important to consider ethics, bias and fair representation.

Adding details such as style, setting and mood in a prompt helps you to control and refine image outputs.

I hope you've enjoyed today's lesson, and I hope you'll join me again soon, bye.