Understand the scenario
Start with the need, then judge the technology.
For May and November 2027. This introductory lesson is shared SL and HL content.
01 | MEET VISIONARY STUDIOS
Imagine being asked to create an advertising campaign before a client has decided what its imaginary world should look like. A designer might explore several settings, colours and characters before one direction is approved. Each image must communicate the right idea, and the images must work together.
Visionary Studios is the creative design company in the 2027 case study. It is evaluating generative AI for advertising campaigns, concept art and digital media. Its goal is to streamline the production of high-quality images. Streamline means making a process more efficient, for example by reducing repeated work or shortening the route from an idea to an approved image.
The company is evaluating possible approaches. The booklet does not say that it has already chosen a winning model or that AI will replace its designers.
02 | WHAT IS THE ACTUAL PROBLEM?
A problem is the need that motivates a change. Here, the challenge is to produce suitable images efficiently while maintaining the quality and control expected by clients. Simply producing more pictures would not solve that problem if most pictures could not be used.
Consider a model that makes an attractive image in seconds but changes the main character in every output. It may be useful for exploring ideas, yet unsuitable for a campaign requiring the same character throughout. The value of a system depends on whether it meets the company’s needs.
This is our teaching summary of the scenario, rather than an additional fact supplied by the booklet.
03 | KEEP FACTS AND EXAMPLES SEPARATE
A case study answer should use the information provided without inventing evidence. We know that Visionary Studios is evaluating image generation models. We do not know its exact budget, staff numbers, computer specifications or current production times.
Throughout these lessons, we will use a fictional teaching brief : a client wants a campaign featuring the same illustrated explorer across six settings. This gives us something concrete to test. The explorer, six settings and any numerical targets are invented examples, not details from the IB booklet.
Use language such as “the company could test…” when proposing a sensible action. Do not turn a proposal into a claim that the company has already carried it out.
04 | THREE CREATIVE USES, DIFFERENT NEEDS
Advertising images must communicate a message and support a brand. An attractive output may still fail if it uses the wrong style or makes a product look inaccurate.
Concept art explores how something might look before a final design is selected. Variety can be valuable here: several unusual settings may help an artist discover a promising direction.
Digital media covers images used in digital creative work. An image may need to fit a particular layout or remain coherent with other assets. These are possible practical requirements to investigate, rather than fixed technical specifications given in the case study.
The same model may be helpful for early experimentation but less suitable for tightly controlled final assets.
05 | WHO NEEDS THE SYSTEM TO WORK?
A stakeholder is a person or group affected by a decision. The designers need usable creative control. Clients need images that meet their brief. Managers need a production process the company can support. Audiences need to be able to trust how content is presented.
Artists and other rights holders whose work might enter training datasets also have interests in the decision. People represented in generated images may be affected by biased or exclusionary outputs.
These stakeholder perspectives are analytical extensions of the scenario. They help us ask better questions; they do not mean that the booklet provides interviews with each group.
06 | FROM A GOAL TO A REQUIREMENT
A goal is a broad outcome, such as improving image production. A requirement states what the proposed solution needs to do or achieve. “Make good images” is a starting point, but it is too vague to guide a fair test.
For the explorer campaign, a more useful requirement would be: “The recurring character should retain recognisable facial features and clothing across the requested settings.” It tells us what to inspect.
A constraint limits the available solution. For example, a project might have a spending limit or a delivery deadline. The case study asks whether existing infrastructure can support a model, but it does not specify a particular budget or deadline.
Produce suitable campaign images efficiently.
Keep the recurring character recognisable.
Review the character across several settings.
07 | FUNCTIONAL AND QUALITY REQUIREMENTS
A functional requirement describes a capability: what a system should do. If a designer needs to submit a description and receive candidate images, generating images from text is a functional requirement.
A quality requirement describes how well the system should work, or a condition it must satisfy. Suitable output quality, acceptable generation time and sufficient capacity are examples. These are often called non-functional requirements.
Some needs can be expressed in more than one way. “Allow a designer to select a style” describes a function; “maintain the selected style consistently” describes output quality. Explain your interpretation rather than treating the labels as a substitute for reasoning.
08 | WHAT COUNTS AS A HIGH-QUALITY IMAGE?
Image quality includes more than a large number of pixels. A high-resolution output can contain distorted objects or fail to follow the brief. Photorealism matters for a realistic campaign; it is not automatically the best choice for a deliberately illustrated one.
The booklet highlights resolution, photorealism, stylistic coherence and alignment with project goals. A useful evaluation checks the dimensions relevant to the particular job.
For our teaching brief, reviewers could inspect whether the explorer is recognisable, the requested setting is present and the illustration style is appropriate. Record both strengths and defects so that a visually impressive result does not hide a serious mismatch.
09 | CONSISTENCY IS A SEPARATE CHALLENGE
One excellent image does not prove that a model can create a coherent campaign. Character consistency means retaining a recurring character’s recognisable appearance across images. Style adherence means following the intended artistic style or theme.
Suppose the explorer’s jacket changes colour and the face changes shape between settings. Each image might look polished in isolation, but the set no longer communicates the same character reliably.
The case study identifies embedding-based approaches as a possible way to support consistency. We will explain those later. At this stage, the important requirement is the consistent result, rather than memorising a technique without understanding the problem it addresses.
10 | EFFICIENCY MEANS USABLE WORK
Computational efficiency concerns the resources needed to perform the work. A model may use substantial processing power, memory and time. Visionary Studios needs to consider whether its infrastructure can support training and deployment.
Training changes a model using data. Generation , also called inference in this context, uses a trained model to produce outputs. Using a model does not necessarily mean training it from scratch.
A fast image generator can still create a slow workflow if designers must reject many outputs or spend a long time repairing them. Measure time to an acceptable result as well as time to the first image.
11 | RELIABILITY, FLEXIBILITY AND SCALE
Training stability concerns whether training progresses reliably towards useful results, rather than becoming unstable or producing repetitive or poor outputs. It is different from maintaining a character’s appearance across images.
Flexibility is the ability to support different creative tasks. A model that works well for landscapes may need further evaluation before being used for character-based branded assets.
Scalability concerns supporting increased demand. A process that works for one designer may create delays when many people request images. Students should explain the effect of that growth on resources and delivery, rather than simply saying a system “is scalable”.
12 | ETHICAL REQUIREMENTS BELONG IN THE BRIEF
A technically impressive model can still be unsuitable. Visionary Studios needs to consider how training data is selected, whether outputs reflect unfair biases and how AI involvement will be disclosed.
Dataset curation means deliberately selecting and organising training data. Checking permissions and provenance is part of a responsible evaluation. An image being publicly accessible does not, by itself, establish permission to use it for every purpose.
Bias checks should look for patterns across outputs, not rely on one apparently fair image. Disclosure guidelines should help staff explain AI involvement clearly to clients and audiences. These are decision criteria to investigate; this lesson does not make a legal compliance determination.
13 | BUILD A REQUIREMENTS TABLE
A table connects each requirement to evidence. The targets below are invented classroom examples , not official IB requirements or measured company results. Their purpose is to show how to make vague needs testable.
| Requirement | Evidence to collect | Example success condition |
|---|---|---|
| Generate candidate images from text. | Submit the agreed briefs and inspect outputs. | Each brief produces reviewable image candidates. |
| Keep the character recognisable. | Review face and clothing across the six settings. | All six selected images retain the agreed identifying features. |
| Reduce production time. | Record total time including rejected outputs and editing. | At least 20% less time than a measured baseline, with quality criteria still met. |
| Support the requested style. | Review selected images against an agreed style rubric. | All selected images meet the agreed style descriptors. |
| Support responsible use. | Document provenance and permission checks, bias review and disclosure procedures. | Required checks are completed and unresolved concerns receive review before use. |
A sensible target should be agreed before comparing results. If the company changes its criteria after seeing which model wins, the evaluation can become biased towards a preferred answer.
14 | COMPARE APPROACHES WITHOUT PICKING A WINNER YET
The company is considering different approaches to image generation. Text-to-image generation takes a textual description as guidance. Other conditional approaches can use labels, sketches or segmentation maps. Unconditional generation does not use that kind of specified content guidance for each output.
Diffusion models are a central part of this case study for SL and HL. GANs and hybrid models are additional HL content. Later lessons will explain how they work and why their strengths and weaknesses matter.
These are not all competing categories at the same level. For example, text-to-image describes an input/output use, while diffusion describes a modelling approach. A diffusion system can be used for text-to-image generation.
15 | A FAIR PILOT BEFORE A BIG DECISION
A pilot is a small trial used to gather evidence before wider adoption. Visionary Studios could prepare representative briefs, decide the review criteria and record the conditions used for each system.
Generate several outputs for each brief. A single successful image may be a lucky result. Record generation time, rejected outputs, editing effort and reviewer judgements. Where systems have different settings, document them so the comparison can be interpreted fairly.
A pilot should include difficult cases, such as recurring characters across different settings. It should also consider dataset information and disclosure arrangements. Attractive images alone are insufficient evidence for adoption.
16 | UNDERSTAND TRADE-OFFS
A trade-off occurs when improving one aspect creates a disadvantage elsewhere. A configuration might produce more detailed outputs but take longer. A tightly constrained style might help brand consistency while limiting exploratory variety.
There is no universal winner for every creative task. A recommendation should connect the evidence to the company’s priorities and acknowledge limitations. “Choose the fastest system” is weak if it ignores whether the resulting images can be used.
A defensible conclusion could recommend one approach for concept exploration while requiring further testing before final campaign production. This is a possible outcome, not the conclusion supplied by the booklet.
17 | WRITE AN ANSWER THAT BELONGS TO THIS CASE
Start with a relevant technical point, explain its effect, then connect that effect to Visionary Studios. Use connectives where they express a real relationship.
For evaluate , consider benefits and limitations and reach a justified judgement. For describe , give relevant characteristics. Do not add unsupported company figures to make an answer appear more specific.
18 | YOUR FIRST CASE STUDY TASK
In your book, write a short proposal for the fictional six-image explorer campaign. Begin with the production problem. Identify three stakeholders, state two functional requirements and three quality or constraint requirements, then explain how you would gather evidence.
Finish with one trade-off and one unanswered question that Visionary Studios should investigate. Clearly label any invented target as a proposal.
Reveal a possible starting point
The studio needs a coherent set of six campaign images. Designers could generate candidate images from a brief and request revisions. Reviewers could check recognisable character features and the agreed illustration style across all six outputs. A pilot could record time to an accepted set, including rejected and edited outputs. Before setting a speed target, the studio should establish its current workflow time.
A faster generator may require more manual correction, so generation speed alone should not determine the recommendation.
Read alongside: the official 2027 case study, especially the scenario and evaluation, ethics and challenges sections. This page is independent teaching material, not an official IB publication.
CHECK YOUR UNDERSTANDING | FACT, EXAMPLE OR PROPOSAL?