L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 3 · Lesson 1

How AI Makes Images

From random noise to recognisable pictures — the surprising math behind AI art
If a computer has never "seen" the world, how can it draw a sunset?
🎨
Note for learners & educators This module is written for curious young learners — roughly ages 11 to 16 — but it uses real facts, real events, and real vocabulary that professionals use. You will meet some big words. That is on purpose. By the end you will own them.

In October 2022, an artist named Jason Allen entered a digital art competition at the Colorado State Fair. His entry, a sweeping fantasy painting called Théâtre D'Opéra Spatial, won first place in the digital art category. When people learned he had used an AI image tool called Midjourney to create it, a fierce debate exploded online. Was it really art? Did he cheat? The judges said he had followed the rules — he had disclosed AI involvement. The internet disagreed loudly. That single blue ribbon sparked a worldwide argument about what it means to make a picture.

What is an AI image generator, exactly?

An AI image generator is a computer program that has studied hundreds of millions of photographs, paintings, and illustrations from the internet. By studying so many images, it learned patterns: what a "dog" looks like, what "sunset" colours tend to be, how shadows fall. It stored all that learning as billions of tiny numbers inside itself.

When you type a prompt — a short description like "a golden retriever sitting in autumn leaves, oil painting style" — the AI uses those numbers to build a new image from scratch that matches your words. It is not copying any one picture. It is creating something new using patterns it absorbed from many pictures combined.

The diffusion process — step by step

The most common technique modern AI image tools use is called diffusion. Here is how it works in plain language:

  1. Start with noise. The AI begins with a completely random, blurry image — imagine static on an old television. There is no picture yet, just random dots of colour.
  2. Read the prompt. It reads your text description and uses it as a target — a direction to "steer" toward.
  3. Remove noise gradually. Over about 20 to 50 small steps, the AI nudges the image away from random noise and toward something that matches the prompt. Each step it asks itself: "Does this look more like the description now?"
  4. Refine details. In later steps, fine details like fur texture, brick patterns, and facial features appear as the noise disappears.
  5. Final image. After all steps are done, the result is a finished picture that nobody painted by hand.
🔬 Why "diffusion"?

The name comes from physics. When you drop a drop of ink into water, it slowly spreads out (diffuses) until the water is evenly coloured. AI diffusion runs that process backwards — starting from the evenly-spread "ink" (random noise) and finding the original "drop" (your image). It is a clever mathematical trick.

Key vocabulary for this lesson
PromptThe text description you type to tell the AI what image to create. More specific prompts usually give better results.
Diffusion modelThe type of AI that starts with random noise and gradually refines it into a matching image. Examples: Stable Diffusion, DALL·E, Midjourney.
Training dataThe enormous collection of images and text the AI studied before it could generate anything. The quality of training data shapes the quality of outputs.
Latent spaceThe invisible "map" of concepts inside the AI's memory — where "dog" is near "cat" and far from "skyscraper." The AI navigates this map when building images.
The main tools you will hear about
DALL·E (OpenAI)

Released in 2021, DALL·E was one of the first AI image tools the public could try. DALL·E 3 (2023) is built into ChatGPT. It is very good at following detailed text prompts accurately.

Midjourney

Used by Jason Allen to win the Colorado fair prize. Known for producing painterly, atmospheric images that feel like illustrations. Operates through Discord — you type commands in a chat.

Stable Diffusion

Released open-source in August 2022. Anyone can download and run it on their own computer. This openness made it hugely popular for experimentation and research.

Adobe Firefly

Adobe's built-in AI, launched 2023. Designed specifically for professional creative work and trained only on licensed images to reduce copyright concerns.

💡 Think about it

When Jason Allen typed his prompt into Midjourney, he made hundreds of decisions — about wording, style, composition, and which of many generated images to select and refine. Does that count as creative work? There is no single right answer. What do you think?

Lesson 1 Quiz

How AI Makes Images — check your understanding
1. What is the starting point of the diffusion process?
✓ Correct! Diffusion starts from pure random noise and gradually refines it toward the target image described in your prompt.
✗ Not quite. The diffusion process starts with random noise — like TV static — and works backwards toward a meaningful image. It does not start from any existing picture.
2. What is a "prompt" in AI image generation?
✓ Right! A prompt is your text instruction. The AI reads it and uses it as a target to guide the image-building process.
✗ A prompt is the text description you type — like "a cat in a space helmet, watercolour style." It guides what the AI creates.
3. Which AI image tool did Jason Allen use to win the 2022 Colorado State Fair digital art prize?
✓ Correct! Allen used Midjourney to create Théâtre D'Opéra Spatial, which sparked a global debate about AI and creative authorship.
✗ It was Midjourney — a tool known for painterly, atmospheric images, which Allen used to create the award-winning Théâtre D'Opéra Spatial.
4. What makes Stable Diffusion different from Midjourney and DALL·E?
✓ Exactly right! Stable Diffusion's open-source release in August 2022 allowed anyone to download and run it locally, making it hugely influential for research and experimentation.
✗ The key difference is that Stable Diffusion is open-source — anyone can download it and run it on their own computer, unlike the closed, web-based tools.
5. What is "latent space" in an AI image model?
✓ Well done! Latent space is like an invisible map inside the AI — concepts that are similar (like "dog" and "wolf") sit close together, while very different ideas are far apart.
✗ Latent space is the AI's internal map of concepts — where related ideas like "cat" and "kitten" are stored near each other. The AI navigates this map when building your image.

Lab 1 — Prompt Engineering

Practice writing effective AI image prompts with your AI tutor

Your mission: Master the art of the prompt

The most important skill in AI image creation is writing a clear, specific prompt. A weak prompt gives a boring result. A strong prompt gives something amazing. In this lab you will practise with your AI tutor — ask it to critique your prompts, suggest improvements, or explain why certain words work better than others.

Talk to your AI tutor below. Have at least 3 exchanges to complete the lab.

Try asking: "What makes a great AI image prompt?" or "Can you help me write a prompt for a dragon flying over a futuristic city at night?"
AI Tutor — Prompt Engineering
Lab 1
Hey! Welcome to Lab 1. I'm here to help you get good at writing prompts for AI image generators. Think of a prompt like directions to an artist — the clearer you are, the better they can paint your vision. What image would you like to describe, or do you want me to explain what makes prompts work?
Module 3 · Lesson 2

Style, Influence & Copying

When does AI "inspired by Van Gogh" cross a line — and who decides?
Can a machine copy a style? Can anyone own a style in the first place?

In January 2023, three artists — Sarah Andersen, Kelly McKernan, and Karla Ortiz — filed a class-action lawsuit against the companies behind Stable Diffusion and Midjourney. Their argument: these AI systems had been trained on their artwork without permission and without payment. When users typed "in the style of Sarah Andersen," the AI produced images that looked uncannily like her work. She had never agreed to train any AI. The lawsuit is still ongoing as of 2024 and may reshape how AI art tools are legally allowed to work.

What does "style" mean in art?

In law, you cannot copyright a style. You can copyright a specific painting, drawing, or photograph — the actual work — but not the general way you work. Nobody can own "impressionism" or "comic-book shading." Human artists have always learned by studying and imitating others. Every art student copies masters to develop their own eye.

The complication with AI: when a human studies Van Gogh and then paints something Van Gogh-like, they produce one painting. When an AI studies millions of images — including Van Gogh's — it can produce unlimited Van Gogh-style images instantly, for free, flooding the market and potentially replacing artists who work in a similar style.

⚖️ The legal grey zone

In February 2023, the U.S. Copyright Office ruled that AI-generated images cannot receive copyright protection because they lack a human author. But what about images where a human gave careful prompts and made many creative decisions? The Copyright Office said it would evaluate these "on a case-by-case basis." No final rulebook exists yet. This is genuinely unsettled law.

The "training data" problem

AI image models learned to draw by studying enormous datasets — collections of images scraped from the internet. The best-known dataset is called LAION-5B, which contains 5 billion image-text pairs gathered from public websites. The images include professional photographs, fine art, medical scans, personal photos posted on social media, and much more.

Many artists, photographers, and designers discovered their work was included in LAION-5B without their knowledge. A website called Have I Been Trained? (haveibeentrained.com) lets people search whether their images appear in the dataset. Many professional illustrators found their entire portfolios inside it.

Some AI companies have responded. Adobe Firefly was trained only on Adobe Stock images and public domain art — images where Adobe had licensing rights. Getty Images launched its own AI generator trained exclusively on its licensed photo library. These "licensed-data" approaches represent one answer to the controversy, though critics point out that even licensing deals often do not financially benefit the original artists fairly.

CopyrightA legal right that gives the creator of an original work exclusive control over how it is used and shared. Lasts for the creator's lifetime plus 70 years in most countries.
Training dataThe images (and text labels) used to teach an AI model. The source and legality of training data is now a major legal and ethical debate.
Style mimicryWhen an AI produces images that closely resemble a specific living artist's recognisable style, raising concerns about economic harm to that artist.
Opt-outSome AI tools now allow artists to request their work be removed from training datasets. Critics say opt-out should be opt-in — artists should give permission first.
Different perspectives to consider
🖌️ Artists' view

Many professional artists feel AI companies profited from theft. Years of skill-building were scraped to train systems that now compete with the very artists who made them possible. Requests to opt out place the burden on creators, not companies.

💻 AI companies' view

Companies argue that AI training is "transformative use" — like a student learning from reference books. No single image is directly copied. The AI creates something new. They say copyright law was not written for this situation.

🧑‍🎓 Users' view

Many people — including young students and hobbyists — see AI art tools as democratising creativity. Suddenly anyone can create a beautiful image, not just those who spent years learning to draw. That openness has real value.

🏛️ Lawmakers' view

Governments worldwide are scrambling to catch up. The EU's AI Act (2024) requires transparency about training data. U.S. Congress has held hearings. Clear rules do not exist yet — which is itself a form of harm to everyone involved.

💡 Think about it

If you spent five years developing a unique illustration style, and an AI company scraped your portfolio to train its model without asking you, how would you feel? Now imagine you are a teenager who just used that tool to create your first piece of digital art. Both experiences are real and valid. Good thinking means holding both at once.

Lesson 2 Quiz

Style, Influence & Copying — check your understanding
1. In January 2023, why did artists Sarah Andersen, Kelly McKernan, and Karla Ortiz sue AI companies?
✓ Correct! The artists argued that their work was scraped and used as training data without their consent, which they called a violation of their rights.
✗ The lawsuit was about training data — the artists argued their work had been used to teach the AI models without their permission or any payment to them.
2. Under current copyright law, can an artist copyright their style?
✓ Right! Copyright protects specific original works — a particular painting or drawing — but not a general style, technique, or approach. Nobody can own "impressionism."
✗ Copyright protects specific works, not styles. You can copyright a specific painting, but not the general way you paint. This is one reason AI style mimicry is legally complex.
3. What is LAION-5B?
✓ Exactly! LAION-5B is the massive training dataset that many AI image models learned from. Its contents include professional art, photos, and personal images — scraped without individual consent.
✗ LAION-5B is a training dataset — a collection of 5 billion image-text pairs gathered from the internet and used to teach AI models to generate images. Many artists discovered their work was in it.
4. Which AI image tool was specifically trained on licensed images to reduce copyright concerns?
✓ Correct! Adobe Firefly was trained on Adobe Stock images and public domain content — images Adobe had rights to — as a direct response to concerns about unlicensed training data.
✗ Adobe Firefly was the tool designed with licensed training data. Adobe specifically used its own Stock library and public domain images to avoid the copyright problems that affected other tools.
5. What did the U.S. Copyright Office rule in February 2023 about AI-generated images?
✓ Right! The U.S. Copyright Office ruled that copyright requires a human author, so purely AI-generated images cannot be copyrighted. Cases with significant human creative input are evaluated individually.
✗ The U.S. Copyright Office ruled that AI-generated images cannot be copyrighted because copyright law requires a human author. This leaves the legal status of many AI artworks uncertain.

Lab 2 — Style & Ethics Debate

Explore the ethics of AI style mimicry with your AI tutor

Your mission: Think through the ethics of AI art

Is it okay to ask an AI to create something "in the style of" a living artist? There is no easy answer. In this lab, debate the question with your AI tutor. Present your view, push back on arguments, and see if your position changes. Good thinkers can argue both sides.

Have at least 3 exchanges to complete the lab.

Try starting with: "I think it's fine to use an artist's style as a prompt — is that wrong?" or "Why should AI companies have to pay artists for training data?"
AI Tutor — Ethics of AI Art
Lab 2
Welcome to Lab 2! This is a genuinely difficult ethical debate — smart people disagree. Let's dig into it together. Here's a starter question: When you type "in the style of [artist name]" into an AI tool, do you think you're borrowing inspiration or taking something that isn't yours? What's your gut reaction?
Module 3 · Lesson 3

Deepfakes & Visual Deception

When AI-generated images are used to mislead — what goes wrong and why it matters
If a photograph can now be completely fake, what happens to trust?

In March 2023, a set of images went viral on Twitter showing Donald Trump being dramatically arrested by police officers. The images were entirely AI-generated, created by journalist Eliot Higgins using Midjourney — as an experiment to see how realistic fake news images could become. They were startlingly convincing. Within hours, millions of people had seen them. Many believed them to be real photographs. Higgins posted them with a disclaimer, but screenshots without the disclaimer spread far faster. The experiment demonstrated that the era of trusting photographs as evidence was ending.

What is a deepfake?

A deepfake is an AI-generated image, video, or audio that depicts a real person doing or saying something they never actually did. The word comes from "deep learning" (the AI method) and "fake." Early deepfakes in 2017–2019 required powerful computers and technical skill. By 2023, creating a convincing fake photograph of any public figure required only a browser and a few seconds.

Deepfakes are not all malicious. Filmmakers use similar technology for legitimate purposes — like digitally de-ageing actors in movies, or bringing historical figures to life in documentaries. The technology itself is neutral. What matters is the intent and the context.

Real-world harms — documented cases

Non-consensual intimate imagery: A 2023 report by the Stanford Internet Observatory found that the vast majority of deepfake content online consists of non-consensual intimate images — fake explicit photos of real people (usually women) who never agreed to be depicted this way. This is a form of abuse. Multiple countries have passed laws making it illegal. In the UK, the Online Safety Act 2023 criminalised sharing such images.

Political disinformation: During elections in Slovakia in September 2023, an AI-generated audio clip circulated on social media appearing to show a politician discussing plans to buy votes. The clip was fake. It appeared just 48 hours before election day — too late for fact-checkers to widely debunk it. Researchers at fact-checking organisations said the timing was almost certainly deliberate.

Financial fraud: In February 2024, a finance worker at a multinational company in Hong Kong was tricked into transferring HK$200 million (about £20 million) after attending a video call where every other participant — including the "CFO" — was a deepfake. The worker had initially suspected a phishing email but was reassured by seeing familiar faces on the call. Police confirmed the case.

⚠️ The scale problem

Before AI, creating a convincing fake photograph of a specific person required a professional photographer, a lookalike, and expensive post-production. Now it takes seconds and costs nothing. This is not just a new kind of problem — it is the same problem at a completely different scale. Scale changes everything.

How to spot AI-generated images

AI image detection is improving, but so is AI generation. Here are current warning signs to look for:

  1. Hands and fingers. AI still struggles with hands. Look for too many fingers, fused fingers, or fingers that bend the wrong way. This is the most reliable tell in 2024.
  2. Text in the image. AI-generated signs, labels, and text often contain garbled, nonsense letters — text that looks real from a distance but dissolves into gibberish up close.
  3. Backgrounds. Edges where objects meet backgrounds can blur, warp, or repeat in patterns. Look at ears, hair edges, and where clothing meets the background.
  4. Eyes too perfect. AI eyes often look slightly glassy, unnaturally symmetrical, or reflect light in ways that don't match the scene's lighting.
  5. Check the source. Ask: Where did this image come from? Is there a news organisation's name attached? Can you find it on a verified news site? Reverse-image search using Google Images or TinEye.
💡 The right response is not panic

Visual literacy — the skill of critically evaluating images — has always been important. Photographers and photo editors have manipulated images since photography was invented. What AI changes is the speed and accessibility. The solution is the same as it has always been: slow down, ask questions, and check sources before sharing. These are learnable skills.

DeepfakeAn AI-generated image, video, or audio that realistically depicts a real person doing or saying something they never actually did.
DisinformationFalse information spread deliberately to deceive people. Different from misinformation, which is false information spread without intending to deceive.
Visual literacyThe ability to interpret, evaluate, and think critically about images — including understanding how they can be manipulated or misleading.
Reverse image searchA tool that lets you upload an image to find where else it appears online — useful for checking whether an image has been altered or taken out of context.

Lesson 3 Quiz

Deepfakes & Visual Deception — check your understanding
1. Where does the word "deepfake" come from?
✓ Right! "Deepfake" combines "deep learning" — the type of AI used — with "fake." The term emerged around 2017 when the technology first became widely noticed.
✗ "Deepfake" comes from combining "deep learning" (the AI technique involved) with "fake." It emerged as a word around 2017.
2. In the March 2023 fake Trump arrest images, what was journalist Eliot Higgins trying to demonstrate?
✓ Correct! Higgins created the images as an experiment with clear disclaimers — but the images spread without those disclaimers, vividly proving exactly the point he was making about the danger of AI-generated disinformation.
✗ Higgins was running an experiment to show how realistic AI fake images had become. He labelled them as fake, but the images spread without his disclaimers — proving his point in an uncomfortable way.
3. Which body part is currently the most reliable way to spot AI-generated images?
✓ Exactly! AI image models still struggle significantly with hands — look for too many fingers, fused fingers, or fingers bending in impossible directions. It is currently the most reliable visual tell.
✗ Hands are the most reliable tell. AI image generators consistently struggle with hand anatomy — producing too many fingers, fused digits, or impossible-looking bends. Always check the hands.
4. What happened in the Hong Kong finance fraud case of February 2024?
✓ Right! This was a landmark case — the worker was reassured by seeing "familiar faces" on a video call, not realising every participant was a deepfake. The fraud totalled approximately £20 million.
✗ In February 2024, a Hong Kong finance employee transferred roughly £20 million after a video call where every participant — including a convincing deepfake "CFO" — was AI-generated. It was a landmark fraud case.
5. What is a reverse image search useful for?
✓ Correct! Reverse image search tools like Google Images and TinEye let you upload a picture and find other places it appears — helping you check if an image is old, altered, or being used in a misleading context.
✗ Reverse image search lets you find where an image appears elsewhere online. It's a key fact-checking tool — if a supposedly "new" image actually appeared years ago in a completely different context, that's a strong sign something is wrong.

Lab 3 — Spotting Fake Images

Build your visual literacy skills with your AI tutor

Your mission: Become a visual detective

Visual literacy is one of the most important skills of the 2020s. In this lab, describe images to your AI tutor and practise applying the detection checklist from Lesson 3. Ask the tutor to quiz you, give you scenario puzzles, or explain why certain features are red flags. Challenge yourself to think critically.

Have at least 3 exchanges to complete the lab.

Try asking: "Quiz me on how to spot AI images" or "Describe a fake-looking AI image and I'll tell you which signs give it away" or "Why do AI images get fingers wrong?"
AI Tutor — Visual Literacy
Lab 3
Welcome to Lab 3 — time to become a visual detective! I'm going to help you build the skills to spot AI-generated images. Here's a challenge to start: Imagine you see a dramatic news photo of a celebrity you recognise at a major event. What are the first three things you'd check to decide if it might be AI-generated or manipulated? Give me your best list!
Module 3 · Lesson 4

AI as a Creative Tool

How real artists are using AI as a collaborator — and what that means for the future of creativity
Is AI a threat to human creativity, or the most powerful new creative tool ever invented?

When AI image tools became widely available in 2022, many professional illustrators, designers, and animators were worried they would lose their jobs. But something more complex happened. Studio Refik Anadol, a data and media arts studio based in Los Angeles, used AI not to replace human artists but to create a kind of art that had never existed before — vast living visualisations of cities, ecosystems, and human memories, displayed as enormous moving murals. Their 2023 installation Unsupervised ran at MoMA in New York, where an AI trained on MoMA's entire art collection of 200-plus years generated constantly evolving dreamlike imagery. Hundreds of thousands of people visited. The Guardian called it "the most talked-about artwork in New York." No single human could have made it — and no AI would have made it without human artistic direction.

Three ways artists are using AI right now

Rather than replacing artists, the most interesting work emerging in 2023–2024 involves AI as a collaborator. Here are three documented approaches:

🎬 Film & Animation

Production studios are using AI for storyboarding, concept art, and background generation — speeding up early creative stages. Director Guillermo del Toro has discussed AI as a "sketch pad." The 2023 SAG-AFTRA and WGA strikes in Hollywood partly concerned AI's role; new contracts now include protections for human writers and actors.

🎮 Video Games

Game studio Ubisoft announced an AI tool called Ghostwriter in 2023 that generates first drafts of NPC (non-player character) dialogue. Human writers then edit and refine the output. The studio says it frees writers to focus on higher-level story work. Other studios use AI for texture generation and environment concept art.

🖼️ Fine Art

Artists like Holly Herndon, Beeple, and the collective Obvious (who sold an AI portrait at Christie's for $432,500 in 2018) have made AI central to their practice. Many argue that the creative decisions — what to make, how to frame it, what to select — are entirely human, even when the brush is algorithmic.

📚 Publishing & Illustration

Book cover designers and publishers have begun using AI-assisted illustration tools. This has sparked controversy — the Science Fiction and Fantasy Writers of America (SFWA) updated its rules in 2023 to require disclosure of AI-generated covers. Several magazines banned AI submissions entirely.

What does "human creativity" actually mean?

This is a genuinely hard philosophical question. Creativity has always involved tools — from cave painters using ochre and their fingers, to Renaissance artists using camera obscura to trace proportions, to photographers using darkroom chemical tricks. Every generation of artists uses the technology available to them.

What seems to make human creativity distinctly valuable is intention and meaning. A human artist makes choices because they want to say something — about the world, about themselves, about their culture. AI does not want to say anything. It responds to patterns. The question of whether that distinction matters — and how much — is something society is currently working out in real time.

📌 The jobs question

A 2023 report by Goldman Sachs estimated that AI could automate tasks equivalent to 300 million full-time jobs globally. Creative fields are included. But the same report noted that new types of jobs are created by new technologies — AI image tools have already created demand for "prompt engineers," AI art directors, and AI ethics consultants that did not exist five years ago. History suggests technology transforms work rather than eliminating it, but transitions are genuinely difficult for real people in the middle of them.

Questions worth sitting with

As you complete this module, consider these questions that do not have simple answers:

  1. Who is the author? When a human writes a prompt and an AI generates the image, who made the artwork? The human who directed it? The engineers who built the model? The artists whose work trained it?
  2. What is worth protecting? If AI can replicate any style in seconds, does style itself lose value? Or does authenticity — knowing a human made something — become more valuable, not less?
  3. What should you learn? If you want to be a creative professional, should you learn traditional skills, AI skills, or both? What skills will still matter in ten years?
  4. Who benefits? AI art tools are mostly built by large technology companies. Who profits from the creativity they enable — and is that distribution fair to the artists whose work trained the systems?
💡 Final thought

The fact that these questions are hard is not a reason to avoid AI tools. It is a reason to use them thoughtfully — with awareness of the debates, respect for the artists involved, and your own creative intentions clearly in mind. The best outcomes happen when human judgment and AI capability work together, not when either one operates alone.

Lesson 4 Quiz

AI as a Creative Tool — check your understanding
1. What was notable about Refik Anadol's "Unsupervised" installation at MoMA in 2023?
✓ Correct! Unsupervised was trained on MoMA's entire art collection and generated constantly changing dreamlike imagery — a form of art impossible without both human artistic direction and AI capability working together.
✗ Unsupervised involved an AI trained on MoMA's entire art collection generating continuously evolving images as a living installation. The Guardian called it "the most talked-about artwork in New York."
2. What is Ubisoft's "Ghostwriter" AI tool designed to do?
✓ Right! Ghostwriter creates draft NPC dialogue — but human writers review and refine the output. Ubisoft describes it as freeing writers to focus on higher-level story work rather than repetitive dialogue drafts.
✗ Ghostwriter generates first drafts of NPC (non-player character) dialogue for human writers to then edit. It is designed as a collaboration tool, not a replacement — though the distinction is a live debate in the games industry.
3. How much did the AI portrait by the collective "Obvious" sell for at Christie's auction house in 2018?
✓ Correct! The portrait, titled "Portrait of Edmond de Belamy," sold for $432,500 in October 2018 at Christie's — far above its $7,000–$10,000 estimate. It was a landmark moment demonstrating market interest in AI art.
✗ The Christie's AI portrait sold for $432,500 in 2018 — a price that shocked the art world and demonstrated that collectors were prepared to pay serious money for AI-generated work.
4. According to the lesson, what distinguishes human creativity most from AI image generation?
✓ Well reasoned! The lesson argues that human creativity is marked by intention — artists make choices because they want to express something. AI responds to patterns without any desire to communicate meaning.
✗ The lesson points to intention and meaning as the key distinction. Human artists make choices to say something about the world. AI responds to patterns without wanting to say anything — it has no intentions.
5. Which organisation updated its rules in 2023 to require disclosure of AI-generated book covers?
✓ Correct! The SFWA updated its guidelines in 2023 requiring disclosure when AI tools were used in creating book covers — reflecting the publishing industry's broader effort to set norms around AI in creative work.
✗ The Science Fiction and Fantasy Writers of America (SFWA) updated its rules to require AI disclosure on covers. Several literary magazines went further and banned AI art submissions entirely.

Lab 4 — Your Creative Vision

Plan an AI-assisted artwork with your AI tutor as creative partner

Your mission: Plan an artwork using AI as a tool

You are the artist. Your AI tutor is your creative assistant. In this lab you will plan — in detail — an original artwork or creative project that uses AI image generation as part of the process. Think about: what you want to say, what style suits it, what prompts you would use, and how you would make the final creative decisions yourself.

Have at least 3 exchanges to complete the lab. This is your creative statement — be specific and ambitious.

Start with your idea: "I want to make an artwork about..." then let your tutor help you develop it into a full creative brief with specific prompts and your own artistic vision.
AI Tutor — Creative Planning
Lab 4
Welcome to Lab 4 — this is where you get to be the artist! I'm here as your creative assistant, not the creator. You set the vision; I'll help you shape it. Tell me about something you care about — a memory, a feeling, a place, a social issue, anything — and let's build an AI art project around it. What's on your mind?

Module 3 Test

AI & Visual Art — 15 questions — 80% required to pass
1. What is the first step of the diffusion process in AI image generation?
✓ Correct! The diffusion process begins with pure random noise and gradually refines it toward the target image.
✗ The diffusion process starts with random noise — like TV static — and works step by step toward the target image described in the prompt.
2. Which AI image tool operates through Discord chat commands?
✓ Right! Midjourney operates via Discord — users type prompt commands in a chat interface.
✗ Midjourney operates through Discord. Users type prompt commands in the Discord chat interface to generate images.
3. Jason Allen's prize-winning AI artwork was called:
✓ Correct! Théâtre D'Opéra Spatial won first place at the Colorado State Fair digital art competition in 2022.
✗ Jason Allen's work was called Théâtre D'Opéra Spatial. It won the 2022 Colorado State Fair digital art prize, triggering a major debate about AI art.
4. "Latent space" refers to:
✓ Right! Latent space is the AI's internal concept map — related ideas cluster together, allowing the AI to navigate from "cat" toward "kitten" or "fur" when building images.
✗ Latent space is the AI's internal map of concepts — where similar ideas sit close together. It's the invisible structure the AI navigates when generating images.
5. The artists who sued AI image companies in January 2023 were suing over:
✓ Correct! Sarah Andersen, Kelly McKernan, and Karla Ortiz sued because their artwork was used as training data without consent or compensation.
✗ The lawsuit centred on training data — the artists' work had been scraped and used to teach AI models without their permission or any payment.
6. LAION-5B is:
✓ Right! LAION-5B is the massive image-text training dataset — containing 5 billion pairs — that many AI image models were trained on.
✗ LAION-5B is a training dataset containing 5 billion image-text pairs scraped from the internet. Many artists discovered their work was included without their knowledge.
7. The U.S. Copyright Office ruled that AI-generated images:
✓ Correct! Copyright requires a human author. Purely AI-generated images fall outside copyright protection under current U.S. law.
✗ The U.S. Copyright Office ruled that AI-generated images cannot be copyrighted — copyright law requires a human author, which AI images lack.
8. Eliot Higgins' fake Trump arrest images (March 2023) demonstrated:
✓ Right! Higgins created the images as a labelled experiment, but they spread without labels — vividly proving how quickly convincing AI disinformation can go viral.
✗ Higgins' experiment showed how realistic and viral AI fake images had become. Even with a disclaimer, the images spread without it — proving the danger of AI disinformation at scale.
9. The most reliable visual sign of an AI-generated image (as of 2024) is:
✓ Correct! Hands remain the most reliable tell — AI consistently struggles with finger anatomy, producing too many, fused, or impossible-looking fingers.
✗ Hands and fingers are the most reliable sign. AI struggles with hand anatomy — look for extra fingers, fused digits, or fingers bending the wrong way.
10. The Hong Kong finance fraud of February 2024 involved:
✓ Right! The worker was convinced by a video call featuring deepfake versions of the "CFO" and colleagues. The total loss was approximately HK$200 million (about £20 million).
✗ A finance worker transferred roughly £20 million after a video call where every participant — including the apparent CFO — was a deepfake. It became one of the most significant deepfake fraud cases ever documented.
11. Adobe Firefly's approach to the training data controversy was to:
✓ Correct! Adobe Firefly was trained exclusively on Adobe Stock and public domain images — a direct response to the copyright concerns raised about other tools.
✗ Adobe Firefly trained only on Adobe Stock and public domain images where Adobe had licensing rights — a deliberate response to the training data controversy affecting other AI image tools.
12. Refik Anadol's "Unsupervised" at MoMA was notable because:
✓ Right! Unsupervised trained on MoMA's full collection history and generated continuously changing dreamlike imagery — a new form of artwork only possible through AI-human collaboration.
✗ Unsupervised involved AI trained on MoMA's entire art collection generating continuously evolving dreamlike imagery — a new kind of artwork that neither a human alone nor AI alone could have created.
13. What does the term "disinformation" specifically mean?
✓ Correct! Disinformation is false information spread with the deliberate intent to deceive — distinct from misinformation, which is false information shared without intending to mislead.
✗ Disinformation specifically means false information spread deliberately to deceive. Misinformation is false information spread without that intent. The distinction matters when thinking about AI-generated fake images.
14. According to the lesson on AI as a creative tool, what most distinguishes human creativity from AI image generation?
✓ Right! Intention and meaning are the key distinction. Human artists make choices because they want to express something; AI responds to patterns without any desire to communicate.
✗ The lesson identifies intention as the key distinction. Humans create to express something — AI responds to statistical patterns without any communicative intent. Whether that distinction matters ethically is for society to work out.
15. The Science Fiction and Fantasy Writers of America (SFWA) updated its rules in 2023 to:
✓ Correct! SFWA's rule update required transparency — readers and publishers should know when AI was used in creating cover art. Several magazines went further and banned AI submissions entirely.
✗ SFWA updated its rules to require disclosure of AI-generated book covers — a transparency measure so readers and publishers know when AI tools played a role in the artwork.