IB COMPUTER SCIENCE | 2027 CASE STUDY | SL & HL

End of case study review

Check your knowledge, apply it to production decisions and write supported answers about Visionary Studios.

START HERE | INDEPENDENT REVIEW

This is an original classroom review, not an official IB examination or markscheme. Attempt the tasks without notes, then check your answers and record specific improvements.

A: scored single-answer knowledge checks. B: scenario-based application and calculations. C: written communication using command words. Finish with a personal review plan.

Shared tasks are suitable for SL and HL. HL students should also complete the marked additional tasks. The familiar 2, 2, 4, 4, 6 and 12-mark written sequence is used as classroom practice , not a confirmed 2027 paper structure. The case study is assessed in Paper 1 at both levels from 2027; use current official specimens and instructions.

All trial figures and additional scenarios here are fictional. Do not quote them as findings from the official case-study booklet.

A | KNOWLEDGE CHECK

Each question has exactly one correct answer and earns one point. Choose your pathway above before scoring. Shared pathway: 14 points; shared plus HL: 20 points.

1. What distinguishes training from ordinary inference?
2. Which input directly supplies region-category layout?
3. What does unconditional generation mean?
4. What happens during diffusion sampling?
5. How are CNN and DDPM related?
6. What is character consistency?
7. Why might short generation time fail to reduce delivery time?
8. What does public access to an image establish by itself?
9. What can equal representation counts hide?
10. What is useful AI disclosure?
11. What makes an evaluation criterion useful?
12. Does using the same seed number across generators guarantee equivalent randomness?
13. What should a fair comparison record?
14. What makes a recommendation defensible?
15. HL: What is the discriminator’s training role?
16. HL: What is mode collapse?
17. HL: What does a VAE encoder represent?
18. HL: What characterises a normalizing flow?
19. HL: Why evaluate a hybrid against a simpler baseline?
20. HL: Which comparison is justified?

B1 | EVALUATE A PRODUCTION PILOT

Fictional trial: same 10 briefs and four candidates per brief
Measure A B
Candidates 40 40
Accepted candidates 24 30
Total production time 240 minutes 330 minutes

In your book: Calculate both acceptance percentages, their difference in percentage points and time per accepted asset. Explain why the data do not prove which workflow is best for recurring-character campaigns. Suggest one additional test.

Reveal checks and reasoning

A: 60% accepted and 10 minutes per accepted asset. B: 75% and 11 minutes. B is 15 percentage points higher on acceptance, while A is quicker per accepted asset. Aggregate acceptance does not establish identity consistency or compare quality among accepted outputs. Test agreed identity features across representative linked scenes and inspect results by brief.

B2 | REPAIR AN UNFAIR COMPARISON

A designer gives Tool A 30 attempts and an hour of editing. Tool B receives three attempts with no editing. Only the best A image and worst B image appear in the report. The report declares that A is the better generator.

Identify three weaknesses. Construct an improved comparison, specifying the question, effort budget, review method and records.

Reveal possible improvements

Attempts and editing differ; selected examples are cherry-picked. Define whether the test compares generators or complete workflows. Give equivalent requirements and a justified comparable budget, record all attempts and versions, repeat representative briefs and apply one rubric. Hide source labels where practicable and score independently. A practical workflow comparison may allow tool-appropriate prompts under a recorded policy.

B3 | MAKE RESPONSIBLE APPROVAL DECISIONS

A proposed campaign uses untraced web images as references. Generated workplace scenes show varied people but always assign leadership to the same group. The delivery note says “fully verified” although only colour balance was reviewed.

Explain a separate concern about sources, fairness and transparency. Suggest a suitable action for each, stating what the action cannot guarantee.

Reveal possible response

Web availability does not establish a suitable basis for each intended use; investigate provenance, terms and relevant context. Equal appearances hide role stereotypes; review portrayal and test relevant changes. “Fully verified” overstates the review; replace it with accurate scope and communicate unresolved issues. Records, review and disclosure address different concerns and none guarantees universal legal clearance, fairness or accuracy.

B4 | TRACE CONDITIONING THROUGH A WORKFLOW

A client provides a character sketch, labelled regions for a scene and a text description. The team claims that any one of these guarantees an identical result.

Compare the guidance each input supplies. Explain why consistency still needs testing across outputs.

Reveal possible response

Text communicates semantic and visual requirements; a sketch can constrain structure or layout; a segmentation map assigns region categories. Their effects depend on the supported workflow and model. They guide different features and do not guarantee identical output or preserved identity. Review agreed features across relevant scenes and record failures and corrective effort.

B5 | EVALUATE A HYBRID PILOT | HL ONLY

A team proposes a hybrid with 120 staff-minutes of setup and an estimated saving of 2 staff-minutes per accepted asset compared with its baseline. Its report claims that combining models guarantees better results.

Calculate the simplified time break-even point. Explain two limitations of the claim and Suggest a useful component comparison.

Reveal possible response

120 ÷ 2 = 60 accepted assets, assuming constant savings and no other differing costs. Added integration and maintenance may change the result, and combining components does not guarantee quality or consistency. Compare the full workflow with an appropriate simpler baseline and, where feasible, remove or vary one component under comparable tests to examine its contribution.

C | WRITTEN COMMUNICATION

Write in your book before revealing guidance. Identify: name precisely. Define: give a precise meaning. Describe: give relevant features or stages. Explain: show why or how through causal links. Discuss: develop relevant perspectives and a considered synthesis.

Use because, resulting in, leading to and therefore when they express a sound relationship. Apply ideas through company consequences, not simply by adding the studio’s name.

The shared sequence below totals 30 practice marks . It develops writing skills at both levels; the 12-mark task does not establish a 12-mark SL examination question. Mark indications and sample guidance are independent classroom suggestions, not official marking instructions or a one-sentence-per-mark rule.

C1 | IDENTIFY [2]

Identify two output characteristics relevant to the studio’s evaluation.

Reveal illustrative answer and review guidance

Style adherence and preservation of required character identity across related images. Check for two distinct relevant characteristics; accept appropriate alternatives.

C2 | DEFINE [2]

Define dataset provenance.

Reveal illustrative answer and review guidance

Information about the origin and processing history of dataset items, such as sources and transformations. Look for meaning and distinguishing detail, rather than an example alone.

C3 | EXPLAIN [4]

Explain why a low generation latency does not necessarily reduce the studio’s final delivery time.

Reveal illustrative answer and review guidance

Fast candidates may fail the brief, requiring more generation and review cycles. Selected images may also need edits before client approval. Therefore total production time can remain high. Look for developed causal links and relevant production consequences.

C4 | DESCRIBE [4]

Describe image generation through iterative denoising in a diffusion-based system.

Reveal illustrative answer and review guidance

Start from a noisy sample. A trained network supplies information used to update it through a sampling schedule, with conditioning where applicable. Repeated updates develop image structure. Convert to the displayable output, decoding if using a latent representation. Do not claim retrieval of a hidden original or retraining on each prompt.

C5 | DISCUSS [6]

Discuss the benefits and limitations of human review of generated campaign images.

Reveal illustrative answer and review guidance

Review can identify deviations from a brief and inappropriate portrayal, supporting approval through contextual judgement. It costs time and reviewers can disagree or introduce bias. Visual checks do not resolve every source or disclosure issue. Define criteria and responsibility, use relevant expertise and escalate concerns. Develop the competing arguments, rather than list them.

C6 | DISCUSS [12]

Discuss whether Visionary Studios should make generative AI its main method for final advertising production.

Reveal illustrative answer and review guidance

Generative AI can offer alternative compositions early, helping clients compare ideas before detailed production. However, generating many candidates is useful only if they support the brief; unsuitable images increase review effort.

Final advertisements need accurate relevant details and consistency. Identity changes across scenes may require editing and regeneration, so the studio should measure complete effort per accepted asset against a suitable baseline. A quick demonstration is insufficient evidence of reliable campaign production.

A generation workflow can support imaginative scenes, while human-led production may offer more precise control for some commissions. Combining methods can preserve useful exploration, but handover and editing add work. Tool choice should therefore follow task requirements.

Sources, portrayal and disclosure affect creators, audiences and clients. Document relevant evidence, review roles and representation, and explain AI contributions accurately. These safeguards take staff time and do not eliminate every uncertainty, but performance benefits do not automatically override essential approval conditions.

Computing resources, provider dependence and staff training also affect adoption. Trial representative briefs, retain failed attempts and record setup and ongoing effort. The fictional pilot on this page may be used if explicitly identified as classroom evidence, but must not be called a booklet finding.

I would recommend staged adoption for suitable tasks rather than immediate replacement of every final-production method. Expand only when evidence shows required quality and consistency with useful overall benefits and completed approval checks. Retain alternatives and review the decision when tools or requirements change.

This is one defensible direction. A different judgement can be strong if developed with accurate technical reasoning, case consequences, relevant alternatives and a supported conclusion.

C7 | TECHNICAL DISCUSSION | HL ONLY

Discuss the suitability of a GAN-based workflow compared with a diffusion-based workflow for the studio. [12 practice marks]

Reveal an answer plan

Explain adversarial training and iterative diffusion sampling accurately, then connect the implications to output quality, diversity, generation effort and intended control. Discuss mode collapse and training instability without assuming all GANs fail or all diffusion systems are superior. Compare actual systems under equivalent briefs, acceptance conditions and budgets. Consider setup, editing, source and approval requirements. Reach a task-specific conclusion and state what evidence would justify it.

A full answer must develop these relationships. This outline is not a complete 12-mark response or official marking scheme.

FINISH | SELF-ASSESS AND PLAN

Keep the automatic knowledge score separate from your written work. The score is a diagnostic, not an IB grade or predicted examination result.

Check your written response Evidence to look for
Accuracy Correct distinctions and mechanisms
Command word The requested type of response
Application Specific consequences for the studio
Development Reasons linked to effects, not repeated claims
Discussion Relevant perspectives, limitations and supported synthesis

Choose one knowledge gap, one application gap and one writing improvement. Record an action for each in your book, then revisit that material and attempt the question again without the sample answer.

Return to revision Worked writing guidance Examination practice