
The IB Computer Science case study 2027 asks a practical question: how should a design company choose generative AI technology for producing images? You must explain machine learning, technical trade-offs and risks.
The case study focuses on Generative AI for image creation, with the subtitle a diffuse vision, and applies to the May and November 2027 examination sessions. Both Standard Level (SL) and Higher Level (HL) students study it, but their expected technical depth differs.
This guide covers the official scenario, key concepts, generative AI architectures, ethical challenges and original practice answers to help you prepare for Paper 1.
What Has Changed in IB Computer Science for 2027?
The IB's official Computer Science curriculum update confirms that the revised course is first assessed in May 2027. The case study is now part of Paper 1 for both SL and HL.
Under the previous syllabus, the case study was assessed through a separate HL Paper 3. That paper has been removed.
| 2027 assessment detail | Standard Level | Higher Level |
|---|---|---|
| Case study examination | Paper 1 | Paper 1 |
| Paper 1 duration | 1 hour 15 minutes | 2 hours |
| Paper 1 contribution to final grade | 35% | 40% |
| Recommended case study teaching time | 15 hours | 30 hours |
The IB Computer Science subject brief for first assessment 2027 specifies three case study questions in Paper 1. It does not provide a fixed mark allocation for each question in the public brief, so avoid relying on unverified marking predictions.
The 2027 Case Study Scenario: Visionary Studios
The IB case study centers on Visionary Studios, a fictional creative design company considering generative AI for advertising campaigns, concept art and digital media.
The company wants high-quality images but must decide whether the benefits outweigh computational, operational and ethical challenges.
Attractive images alone are not enough. Brand consistency, processing demands, training stability, image control and legal risks all affect the suitability of a model.
A publicly accessible 2027 case study reference page reproduces the terminology from the booklet. Students and teachers should still confirm the complete IB-issued document and examination scope with their coordinator.
How Does Generative AI Create Images?
Generative artificial intelligence uses learned statistical patterns to produce new content. Unlike a conventional graphics tool driven by explicit drawing commands, a generative model can create new visual arrangements.
Three approaches are particularly relevant to the case study:
- Text-to-image generation: A text prompt describes a desired scene or visual style. The system uses that information to guide image generation.
- Conditional generation: Additional information, such as a category label, sketch or segmentation map, influences the generated output. A segmentation map identifies regions belonging to different object classes.
- Unconditional generation: A model generates images without a specific prompt or class label, drawing on patterns learned during training.
Image-to-image translation is another relevant technique. For example, a model can transform a rough sketch into a more realistic rendering while retaining important features of the original image.
Machine Learning: Training vs Image Generation
During training, a machine learning model learns parameters from a collection of images, often accompanied by text descriptions.
Developers prepare and filter the data, define a training objective and update model weights to reduce error. Training quality depends on data coverage, labels, architecture and optimization.
During inference, the trained model uses its learned parameters to generate an image. It is not normally retrained from scratch for every user prompt.
Dataset curation is particularly important for Visionary Studios.
Inadequate training data can limit the range of visual styles represented, reproduce stereotypes or create legal uncertainty. Simply increasing dataset size does not guarantee fairer or more appropriate outputs.
Diffusion Models and Denoising Explained
Diffusion models are the main technical foundation for both SL and HL.
The case study discusses denoising diffusion probabilistic models (DDPMs), which learn to reverse a process that gradually corrupts images with noise.
During training, different amounts of random noise are added to real images. A neural network learns to predict the noise or the information needed to remove it. Repeated training updates improve its ability to recover meaningful image structure.
During generation, the model starts with random noise and gradually transforms it into an image. In conditional generation, a text prompt or another input guides this process.
The case study identifies convolutional neural networks (CNNs) as a common denoising architecture. CNNs learn visual features from spatial patterns within images.
The Hugging Face Diffusion Course provides further explanations of model training and image sampling.
The central trade-off: Repeated denoising requires computational resources and processing time.
Visionary Studios could investigate faster sampling methods, different hardware or reduced image resolution. However, it must test whether these changes affect output quality or consistency.
Fewer denoising steps do not automatically preserve the same image quality.
GANs and Hybrid Models: Higher Level Only
The official case study explicitly reserves generative adversarial networks (GANs) and hybrid models for Higher Level students.
SL students are not expected to know these sections.
A GAN trains two neural networks together:
- The generator transforms a random input vector into synthetic images.
- The discriminator learns to distinguish generated images from real training examples.
Feedback from this competitive process helps improve the generator.
However, GAN training can become unstable. One important problem is mode collapse, where the generator repeatedly produces a limited variety of images instead of representing the full diversity of the training data.
A hybrid model combines different generative approaches to balance their strengths.
Two additional architectures are relevant:
- Variational autoencoder (VAE): Learns a structured, lower-dimensional latent representation of images, supporting generation and editing.
- Flow-based model: Learns an invertible transformation between data and a simpler distribution, supporting image generation and mappings between representations.
HL students need deeper understanding of diffusion, GANs and hybrid architectures. The case study specifies that a broader understanding of VAEs and flow-based models is sufficient.
How Should Visionary Studios Evaluate a Generative AI Model?
A strong case study answer evaluates technology against the company's requirements, rather than declaring one architecture universally superior.
| Evaluation factor | Key question |
|---|---|
| Output quality | Are generated images clear, convincing and appropriate for the campaign? |
| Style adherence | Do images match the requested visual theme and brand identity? |
| Character consistency | Does a recurring character retain recognizable features across multiple images? |
| Computational efficiency | What processing time, hardware and resources are required? |
| Training stability | Can the model train reliably while maintaining output diversity? |
| Scalability | Can the system support increasing production demands? |
The case study mentions embedding-based approaches for maintaining character consistency.
These approaches use learned numerical representations to help retain recognizable character features across generated outputs.
Consider a hypothetical advertising campaign featuring one recurring character. A fast model might be unsuitable if it repeatedly changes that character's appearance.
Conversely, a concept art project might value generation speed and visual variety more highly.
A convincing recommendation should explain how the importance of each requirement changes the preferred solution.
Ethical Risks: Copyright, Bias and Transparency
The 2027 case study explicitly addresses intellectual property, dataset curation, bias mitigation and transparency.
Intellectual property: Visionary Studios should document training data sources, review permissions and licenses, and consider applicable copyright law. The legal treatment of AI training and generated outputs depends on jurisdiction and circumstances.
Bias and representation: If training data underrepresents certain groups, locations or professions, generated advertising images may reproduce exclusionary patterns. Dataset review, varied test prompts and human evaluation can help identify and mitigate these risks.
Transparency: Clients and audiences may need to know when images are AI-generated. Disclosure policies can protect trust, but should also be supported by checks for misleading or unauthorized content.
The NIST Generative AI Risk Management Profile provides an authoritative framework for evaluating risks throughout an AI system's life cycle.
Connect the Case Study to IB Assessment Objectives
The 2027 Computer Science subject brief identifies four assessment objectives.
- Knowledge and understanding: Define terms such as denoising, conditional generation and mode collapse accurately.
- Application: Explain how a technique could address the practical requirements of Visionary Studios.
- Analysis and evaluation: Compare quality, computational cost, consistency and risk before defending a recommendation.
- Computational thinking: Break a problem into constraints, identify measurable success criteria and propose testing methods.
These are course-wide objectives, not a prediction that each will appear as a separate case study question.
Original Paper 1 Practice Questions and Sample Answers
The following questions are illustrative revision exercises, not official IB questions or marked exemplars.
Question 1: Explain why diffusion image generation can be computationally demanding.
Sample answer:
A diffusion model generates an image through multiple denoising steps. Each step requires neural network calculations, increasing processing time and resource usage. Visionary Studios could investigate faster sampling methods but would need to evaluate whether reducing computational demands affects the quality of generated images.
Question 2: Discuss one reason dataset curation is important to Visionary Studios.
Sample answer:
Training images influence the patterns a generative model learns. An unbalanced dataset could result in stereotyped or limited visual representations that are unsuitable for advertising audiences. Reviewing dataset composition and testing outputs across varied prompts would help identify problems, although these measures cannot guarantee complete fairness.
Question 3, HL only: Evaluate a trade-off between GANs and diffusion models.
Sample answer:
A GAN can generate images using a trained generator, potentially supporting efficient inference. However, its adversarial training can be unstable and may experience mode collapse. Diffusion models generally require iterative denoising, increasing computational demands while providing another approach to generating high-quality, varied images. Visionary Studios should compare output quality, generation time and training stability before selecting an architecture.
IB Computer Science 2027 Case Study Revision Checklist
Read the official case study booklet first. For every important term, prepare a definition, an explanation of the underlying mechanism and an application to Visionary Studios.
Focus your revision on these tasks:
- Explain conditional, unconditional and text-to-image generation.
- Draw and describe diffusion training and inference separately.
- Compare image quality, computational efficiency and consistency.
- Explain how dataset curation influences output quality and bias.
- Evaluate legal, ethical and transparency considerations.
- For HL, explain GAN training, mode collapse and hybrid models.
Practise explanation and evaluation rather than memorizing definitions alone. Support recommendations with technical reasoning and trade-offs.
Frequently Asked Questions
Is the 2027 Computer Science case study only for HL?
No. Both SL and HL students study it for Paper 1. However, GANs and hybrid models are HL-only material.
Is there still a Computer Science Paper 3?
No. The revised assessment uses Paper 1 and Paper 2, with the case study assessed in Paper 1.
Do students need to train an image generator?
The published subject brief does not identify building or training a generative AI model as a Paper 1 case study requirement. Students should focus on understanding and applying the prescribed concepts.
Will the exam require a specific model recommendation?
There is no publicly verified fixed question wording. Prepare to justify different model choices using relevant technical and ethical trade-offs rather than memorizing one preferred answer.
Final Preparation Advice
Combine accurate machine learning explanations with practical judgment. Understand how each model works, identify what Visionary Studios needs and evaluate alternatives using evidence and limitations.
Use the official IB Computer Science update and your school's 2027 case study booklet as the final authority on assessment scope.
Table of contents
Share
Related reading
- IB Computer Science IA 2027: Criteria, Format and ExamplesUnderstand the IB Computer Science IA 2027 rubric, 2,000-word report, video and code appendix. Explore a worked project example and practical evidence tips.
- IB Computer Science: What You Really Study and BuildIs IB Computer Science right for you? Explore the 2027 syllabus, SL and HL papers, coding expectations and IA requirements before choosing your DP subjects.
- IB Computer Science 2027: Paper 1 vs Paper 2 ExplainedUnderstand IB Computer Science Paper 1 vs Paper 2 for 2027, including SL and HL timings, marks, Theme A, Theme B, case studies and revision strategies.
- IB Computer Science 2027 Paper 2: Python or Java?Prepare for IB Computer Science Paper 2 2027 with Python and Java examples, algorithm tracing, debugging, OOP guidance and a practical exam revision plan.
- Can IB Math AI HL Get You Into Computer Science in 2027?Can IB Math AI HL get you into Computer Science? Compare 2027 university entry rules at Oxford, UCL, Cambridge and more to choose your subjects confidently.
- IB Global Politics Paper 3: Case Study Strategy GuideMaster Global Politics Paper 3 with a practical case study matrix, reliable sources, six lines of inquiry and original practice questions. Start preparing.
- IB Business Paper 1: Preparing for the Pre-Release CasePrepare for IB Business Paper 1 with a pre-release case study method, stakeholder analysis, business tools and essay strategies. Start revising today.
