The most effective IB case study study strategy is a repeating cycle: read and annotate the booklet, master its Additional Terminology, research each technical topic independently, record useful studies and examples, and practise written responses to every challenge. Reading the booklet once is only orientation. To prepare properly, you must convert its content into knowledge that you can recall, apply to Visionary Studios, and evaluate under timed conditions.
For the May and November 2027 sessions, the case study is “Generative AI for image creation -- a diffuse vision.” It concerns Visionary Studios, a creative design company evaluating generative AI for advertising, concept art, and digital media. This guide explains what SL and HL students need to prepare, how to organize their research, and how to turn that research into effective Paper 1 answers.
What the 2027 case study assesses
Under the Computer Science course first assessed in 2027, the pre-seen case study is included in Paper 1 for both SL and HL. The former HL Paper 3 has been removed. The IB’s official Computer Science curriculum update confirms that Paper 1 now combines Theme A with the case study.
Paper 1 lasts 1 hour 15 minutes at SL and 2 hours at HL. It contributes 35% of the final grade at SL and 40% at HL. Within that paper, the case-study section is worth 12 marks at SL and 24 marks at HL, so it deserves planned preparation rather than a brief review before the examination.
RequirementSLHLCase study assessed in Paper 1YesYesPaper 1 duration1 hour 15 minutes2 hoursPaper 1 weighting35%40%Case-study marks1224Challenges to prepare24GANs and hybrid models requiredNoYes
The official booklet is distributed through IB schools and the Programme Resource Centre. Use the copy supplied by your teacher or coordinator as the definitive version rather than relying on summaries, which may omit level distinctions or important wording.
Understand the 2027 scenario before researching it
Visionary Studios is considering several approaches to image generation. Both SL and HL students need an in-depth understanding of diffusion models, including noise injection, iterative denoising, DDPMs, and the role of a convolutional neural network as a denoiser. They must also understand conditional, unconditional, and text-to-image generation.
The company evaluates models using five broad factors:
-
Output quality, including resolution, realism, and stylistic coherence
-
Computational efficiency, including processing time and infrastructure demands
-
Training stability, meaning whether training reliably produces useful outputs
-
Flexibility and scalability across projects and workloads
-
Consistency, including recurring characters and adherence to a required style
Both levels also examine dataset curation, intellectual property, bias, fairness, and disclosure of AI-generated content. HL students additionally study GANs, the relationship between generator and discriminator networks, mode collapse, VAEs, flow-based models, and hybrid architectures.
This distinction matters. SL students should not spend limited revision time mastering GAN mathematics or hybrid architectures, because the booklet explicitly identifies these as HL material. HL students, however, need enough understanding to compare architectures rather than merely define them.
The repeatable five-stage study cycle
A strong approach revisits the case study several times, with each cycle producing a new revision resource. A two-week rotation works well, but the precise schedule matters less than completing every stage.
1. Read the booklet closely
On the first reading, identify the scenario, stakeholders, technologies, objectives, and constraints. On the second, annotate every sentence that describes how a model works, why Visionary Studios might use it, or what could prevent successful implementation.
Use a simple annotation code:
-
T for technical mechanism
-
B for benefit
-
L for limitation
-
E for ethical or legal issue
-
V for a detail specific to Visionary Studios
-
? for something requiring independent research
Then reconstruct important processes from memory. For a diffusion model, for example, you should be able to explain the movement from random noise through repeated denoising steps to a coherent image, and connect that iterative process to computational cost. If you can only recognize the explanation when reading it, you do not yet know it well enough for an examination.
2. Learn the Additional Terminology
The Additional Terminology list is a minimum vocabulary set, not an optional glossary. SL and HL share terms such as diffusion model, DDPM, noise injection, denoising, CNN, dataset curation, bias mitigation, and character consistency. HL has additional terms including GAN, generator, discriminator, latent space, mode collapse, and hybrid model.
For each term, prepare three levels of knowledge:
LevelWhat you should be able to doExample for “denoising”DefinitionGive its precise meaningRemoving predicted noise from dataMechanismExplain how or why it worksA trained neural network estimates noise over repeated stepsApplicationConnect it to the scenarioMore denoising steps may improve output but increase generation time
Use active recall rather than rereading. The RevisionDojo A4 Machine Learning flashcards can support syllabus vocabulary, but you should create a separate deck for the case-study-specific list. Include reverse cards that present an application and require you to identify the relevant term.
3. Research the topics independently
Treat every section heading and challenge as a research prompt. The booklet gives the scope, but independent investigation gives you the depth needed to explain trade-offs, propose solutions, and evaluate consequences.
For each topic, answer six questions:
-
How does the technology work?
-
What problem does it solve for Visionary Studios?
-
What technical or organizational limitation does it create?
-
What solution or mitigation could address that limitation?
-
What trade-off would the solution introduce?
-
What real example, study, or organization demonstrates the point?
Prioritize reputable sources such as original research papers, university publications, recognized technical documentation, government guidance, and relevant legal authorities. Do not accept an AI-generated explanation without checking its underlying sources. Jojo AI can help generate questions or identify gaps in your understanding, but verification and final reasoning must remain your own.
The RevisionDojo A4 Machine Learning resources are useful for revising the underlying syllabus before adding case-study-specific research. This prevents a common problem: collecting advanced facts without understanding the foundational concepts that make those facts meaningful.
4. Take notes on cited studies and examples
Create a research table rather than copying paragraphs into a document. Each row should contain one usable piece of evidence.
FieldWhat to recordSourceAuthor, organization, or research groupDatePublication yearFindingOne accurate sentence in your own wordsRelevanceThe case-study challenge it supportsLimitationWhat the evidence does not establishExam useThe argument or counterargument it could support
The IB does not require a formal bibliography in a Paper 1 response. Nevertheless, remembering the organization or study behind a finding can make your evidence more precise and reduce the risk of vague claims. Never invent a statistic, publication, or quotation when you cannot recall it accurately.
Aim for a small collection of flexible, well-understood evidence rather than dozens of isolated facts. One example concerning biased image datasets might support arguments about dataset curation, representational fairness, reputational risk, and the need for human review.
5. Practise responses to the Challenges Faced
The challenges indicate where your research must lead. Both levels prepare:
-
Managing iterative denoising and its computational demands while generating photorealistic images.
-
Navigating intellectual property and bias mitigation when using large training datasets.
HL students additionally prepare:
-
Balancing the generator and discriminator in a GAN to produce realistic images.
-
Evaluating how hybrid models could combine the strengths and weaknesses of VAEs, GANs, flow-based models, and diffusion models.
For each challenge, write a one-page planning sheet containing the mechanism, problem, two possible solutions, advantages, limitations, stakeholder effects, and a justified recommendation. Then turn the plan into timed responses using different command terms.
The RevisionDojo Computer Science Questionbank and its machine-learning questions can strengthen the underlying knowledge. Use Computer Science predicted papers later for full-paper timing rather than as a substitute for topic research.
How to structure case-study answers
Short responses should answer the command term directly and use the scenario. If asked to outline a computational difficulty caused by diffusion models, do not write only that they are “expensive.” Explain that repeated denoising requires multiple inference steps, increasing processing time and hardware use when Visionary Studios generates high-resolution campaign images.
For an extended response, use this structure:
-
Position: Establish the decision or issue being evaluated.
-
Technical explanation: Show how the relevant model or process works.
-
Argument: Explain a benefit or proposed solution in the scenario.
-
Counterargument: Examine a limitation, cost, or competing priority.
-
Evidence: Integrate a researched example where relevant.
-
Judgment: Reach a conditional conclusion tied to Visionary Studios.
An effective judgment is rarely “yes” or “no.” It might conclude that diffusion models are appropriate for high-quality concept art if Visionary Studios can accept slower inference and use optimization or suitable cloud infrastructure, but less appropriate where rapid, low-cost generation is the dominant requirement.
Common mistakes that reduce marks
-
Summarizing instead of answering: Retelling the booklet does not address an evaluative question.
-
Giving generic AI points: Every major point should connect to Visionary Studios, image generation, or the stated challenge.
-
Memorizing definitions alone: Definitions help with short questions but cannot replace explanation and evaluation.
-
Ignoring trade-offs: A proposed solution can improve speed while reducing quality, or reduce bias while narrowing the dataset.
-
Using unsupported research claims: An invented statistic is less valuable than an accurate qualitative example.
-
Studying outside the level requirement: SL students should prioritize the two shared challenges; HL students must prepare all four.
-
Practising only untimed plans: Knowledge must eventually be converted into complete answers under examination conditions.
A practical weekly schedule
SessionMain taskOutput1Close readingAnnotated booklet and question list2Terminology recallTested glossary cards3Independent researchThree to five evidence-table entries4Challenge planningOne structured response plan5Timed writingOne short response and one extended response6ReviewError log and revised answer
Repeat the schedule with a different challenge. Every third cycle, complete a mixed Paper 1 practice so that case-study preparation remains integrated with Theme A rather than becoming a separate subject.
Conclusion
Studying the 2027 IB Computer Science case study means repeatedly transforming a pre-seen booklet into examinable understanding. Read it closely, learn every required term, investigate the technical and ethical issues, retain a manageable set of credible studies, and practise applying that knowledge to the Challenges Faced.
SL students should master diffusion models and dataset ethics, while HL students must also evaluate GANs and hybrid approaches. RevisionDojo’s Machine Learning notes and flashcards can reinforce knowledge, while the Questionbank, Jojo AI feedback, and Predicted Papers can help convert that knowledge into precise, timed responses.
