Thing 9

AI image generation

Last reviewed: March 2026 30–45 minutes

Welcome to Phase 3. For the first eight Things, you've been working almost entirely with text: writing prompts, analysing documents, building notebooks, and commissioning research reports. That changes now. AI image generation lets you describe what you want to see, in plain English, and have a machine create it in seconds. No photography, no illustration skills, no stock photo subscription required.

If you've seen AI-generated images online (the viral fake Pope in a puffer jacket, uncanny photorealistic portraits, the slightly off-looking hands that became a running joke in 2023) you already know the results range from impressive to bizarre. What you might not know is how accessible the tools have become. You don't need to install anything, learn any technical skills, or spend any money to start generating images right now.

This Thing is about getting your hands on the tools, understanding what they're good at (and where they fall apart), and developing a feel for how the way you describe an image shapes what comes back. If you enjoyed the prompt engineering work in Thing 3, you're going to find that image prompting takes those same skills in a completely different, and often more satisfying, direction.


How AI image generation works

An abstract illustration representing AI image generation, with colourful shapes emerging from a text prompt
AI image generators create new images from text descriptions, guided by patterns learned from millions of existing images.

You don't need a deep technical understanding to use these tools effectively, but a rough sense of what's happening helps you write better prompts and understand the results.

AI image generators are trained on enormous datasets of images paired with text descriptions. Through this training, they learn the relationships between words and visual concepts: what "sunset" looks like, what "watercolour" means as a style, how "close-up portrait" differs from "wide landscape shot." When you type a prompt, the model isn't searching a library of existing images or stitching photographs together. It's generating something new, pixel by pixel, guided by the patterns it learned during training.

This is why the results can be both impressive and unpredictable. The model has a sophisticated understanding of visual concepts but no real-world experience. It knows what hands generally look like based on millions of images, but it doesn't understand that humans have five fingers, which is why AI-generated hands were notoriously unreliable for a long time (though the latest models have improved). It knows that text on a sign should look like letters, but until recently most models couldn't reliably spell words correctly.

The practical takeaway: the more specific and descriptive your prompt, the more control you have over the output. Vague prompts produce generic images. Detailed prompts produce results that are much closer to what you had in mind.


What's available

The image generation space has expanded rapidly, and there are now several strong options on free tiers. Each tool has a distinct personality: they respond differently to the same prompt, have different strengths, and produce noticeably different visual styles.

For this activity, we'll focus on the three tools with the most accessible free tiers: Ideogram, Bing Image Creator, and one more of your choice.


What makes a good image prompt

Writing prompts for image generation is a different skill from writing prompts for chatbots. With a chatbot, you're giving instructions and context. With an image generator, you're painting a picture in words, and every detail you include (or leave out) shapes the result.

Think about the same elements a photographer or art director would consider: subject, setting, style, lighting, composition, and mood. Compare these two prompts:

A dog in a park

A golden retriever sitting on freshly cut grass in a sunny park, soft afternoon light, shallow depth of field, the dog looking directly at the camera with a slightly tilted head, warm colour palette, photorealistic style

The first prompt will give you an image. It might be fine. But you have almost no control over what you get; it could be any breed, any park, any style, any angle. The second gives the model enough information to produce something much closer to a specific vision.

Some elements that make a noticeable difference:

One thing to keep in mind: the same prompt will produce noticeably different results across different tools. That's useful for this activity. Comparing how different generators interpret the same words is one of the best ways to understand what each tool does well.


Ethics and limitations

Before you dive into the activity, there are a few things worth knowing about how image generation works and what it means to use it.

None of this should stop you from experimenting. Image generation is useful and often fun. But it's worth approaching it with your eyes open.


Resources to explore

Ideogram

Free tier with slow-queue generations. Sign up with email, Google, or Apple. The best option for text rendering in images.

Visit site
Bing Image Creator

Free with a Microsoft account. 15 fast generations per day, unlimited slower generations. Powered by DALL-E 3 and GPT-4o.

Visit site
ChatGPT

Free tier includes 2–3 image generations per day. Conversational interface allows iterative refinement.

Visit site
Leonardo.ai

Generous free tier with daily token allowance. Strong on artistic and stylised images.

Visit site
Adobe Firefly

Limited free credits. Trained on licensed content, so outputs are designed to be safe for commercial use.

Visit site
Prompt engineering for image generation (Tokenized)

Written for Stable Diffusion, but the principles of descriptive prompting transfer well to any image generator.

Read article

Activity: the image generation showdown

30–45 minutes Ideogram, Bing Image Creator, plus one more

You're going to write a single, detailed image prompt and feed it to at least three different tools, then compare the results side by side. This is the image generation equivalent of the chatbot comparison from Thing 2: same input, different tools, and you'll be surprised by how much the outputs vary.

  1. Write your prompt. Choose a subject that interests you: a hobby, a place you know, an imaginary scene, or a practical image you could actually use. Write one detailed prompt that includes subject, setting, style, and mood. For example: "A cosy independent bookshop on a rainy Edinburgh street at dusk, warm light spilling from the windows onto wet cobblestones, a ginger cat sitting in the doorway, watercolour illustration style."
  2. Generate with Ideogram. Go to ideogram.ai and create a free account if you haven't already. Paste your prompt and generate. Save the results and note how closely the images match what you described.
  3. Generate with Bing Image Creator. Go to bing.com/create and sign in with your Microsoft account. Enter the exact same prompt. Compare the style, detail, and mood with Ideogram's output.
  4. Generate with a third tool. Try ChatGPT, Leonardo.ai, Adobe Firefly, or any other free image generator. Use the same prompt again.
  5. Compare and reflect. Lay out all your results side by side and write your evaluation. Which tool came closest to what you described? Which produced the most visually appealing result? If your prompt included text, how did each tool handle the lettering? Which tool would you reach for first if you needed an image for a real purpose?
Privacy reminder: choose personal subjects, hobbies, or fictional scenarios. Avoid images of real people or anything connected to your workplace.

Your output: a document containing your original prompt, the generated images from all three (or more) tools clearly labelled, a written comparison covering prompt accuracy and visual quality, and a brief reflection on what surprised you, what you'd use image generation for, and what limitations you noticed.

Going further

If you have time, try one of these optional extensions:

Why this matters

Image generation is one of the most accessible applications of AI, and beyond the wow factor it has practical uses. Professionals in every sector spend time searching for stock images, commissioning illustrations, mocking up designs, or trying to make a slide deck look less bland. AI image generation won't replace professional designers, but it gives everyone the ability to create custom visuals quickly and for free.

What matters more than knowing the tools exist is understanding how they interpret the same description differently, learning to write prompts that get you closer to what you want, and developing the judgement to know when an AI-generated image is good enough and when it isn't.

The comparison format is important here too. Just as with the chatbot comparison in Thing 2 and the research comparison in Thing 8, putting tools side by side is the fastest way to develop an informed perspective. You'll leave this activity with a much clearer sense of which image generators suit your needs than any review article could give you.


Claim your Open Badge

Your submission should include your prompt, the generated images from at least three different tools, and your written comparison with reflections on quality, accuracy, and potential uses.

Thing 9: AI image generation open badge
Thing 9: AI image generation

Submit your image generation comparison with prompt, results, and reflections as evidence to claim this badge via cred.scot.

Claim now

What's next

Now that you've tried visual generation, Thing 10 moves into a different kind of AI creativity: voice and speech. You'll explore both directions, getting AI to speak (text-to-speech) and getting AI to listen (speech-to-text). If generating an image from a text description felt like magic, wait until you hear a natural-sounding voice read your words back to you in any accent you choose.