IB Computer Science case study 2027

The case study for the May and November 2027 IB Computer Science exams is reported as "Generative AI for image creation: a diffuse vision". It follows a creative company deciding whether, and how, to use AI systems that make images from text prompts.

Key idea: You are not asked to remember the story. You are asked to apply real computer science to it. That means knowing how diffusion models turn noise into a picture, where convolutional neural networks fit in, what goes wrong with training data, and how to weigh quality against speed and cost. HL students are reported to also need generative adversarial networks (GANs) and hybrid systems that combine more than one kind of model.

Everything on these pages is explained in our own words with our own examples. Read your school's copy of the case study alongside it, and use these pages to understand the computer science underneath.

The case study is reported to be examined in Section B of Paper 1, at both SL and HL. Paper 3 is reported to have been removed in the new course.

• SL is reported to prepare the first two challenges, with Section B worth 12 marks.
• HL is reported to prepare all four challenges, with Section B worth 24 marks.
• Questions are likely to include short ones that check you know the terms (define, state, outline), questions that apply them to the company (explain, suggest), and at least one longer question where you weigh up options and reach a conclusion (discuss, evaluate, to what extent).

Key idea: The long evaluative question is where most students lose marks, so practise writing one under time. Every topic page below ends with questions in that style, and the practice page has full-length ones with marking points.

1. Read the case study once for the story, and list every technical term in it.
2. Learn each term well enough to explain it to a friend without notes. Our glossary is a place to start.
3. Work through the topic pages for your level, in order: each one builds on the last.
4. Do the practice questions on paper first, then mark yourself against the marking points. Be strict: a point counts only if it is clearly made.
5. In the last two weeks, write one timed extended answer every few days.
6. Do some research of your own. Students who bring in real systems and real numbers (how many steps a sampler uses, how long a generation takes) write stronger evaluations.

Every practice question on these pages uses our own made-up company, so you practise applying ideas to a context you have not seen, just as in the exam.

Hearth & Oak is an online furniture shop. It photographs each product in a studio, but customers say they cannot picture a sofa in a real room. Hearth & Oak wants to generate realistic room scenes around its products, and to add a "Design my room" box where a customer types a description ("a small bright flat with a green velvet sofa") and gets an image back in a few seconds. It has a small team, a limited budget for cloud graphics processing units (GPUs), and a brand that depends on customers trusting that what they see is what they will get.