Understand the technology
Connect creative capability with production needs.
For May and November 2027. This introductory lesson is shared SL and HL content.
01 | A SYSTEM THAT WORKS TODAY MUST FIT TOMORROW’S WORK
Visionary Studios may use an image generator for a landscape concept today and a character-based advertising campaign tomorrow. As demand grows, several designers may need it at the same time.
Flexibility concerns adapting to different tasks and requirements. Scalability concerns supporting increased demand with appropriate resources. Brand requirements define how creative work should represent a particular organisation or campaign.
The case study connects adaptability, future demand and consistent artistic outputs. These concerns interact, but each needs its own evidence.
02 | FLEXIBILITY MEANS MORE THAN MANY PROMPTS
A system may accept almost any sentence while producing useful results for only a narrow range of briefs. Input freedom is not the same as useful task coverage.
A flexible workflow might support concept exploration, controlled illustrations and revisions from approved references. It may also need suitable output formats and editing options.
Test representative tasks rather than assuming that success on one impressive landscape transfers to every advertising job.
03 | MATCH THE INPUT TO THE CREATIVE NEED
Text can communicate a new idea quickly. A sketch can communicate arrangement. A segmentation map can specify where categories belong. Reference information may help preserve appearance.
Supporting several input types can increase creative control, but only if the system uses them effectively. A long feature list does not prove that they work together.
For the fictional explorer campaign, text may establish a scene while an approved reference supports identity. Those are teaching possibilities, not tools the booklet says the company already uses.
04 | ADAPTATION CAN HAVE A COST
A new style may require prompt changes, reference preparation, editing or further model adaptation. These activities differ in effort and computing needs.
Supplying a reference during inference does not necessarily retrain the model. Fine-tuning involves further learning and introduces additional data and evaluation requirements.
Measure how much work is needed to move between briefs. A system that can eventually produce the desired result may still be impractical if every new client requires extensive specialist preparation.
05 | FLEXIBILITY IS NOT A GUARANTEE OF ACCURACY
A generator might produce realistic and illustrated outputs while altering product geometry or written text. Broad stylistic range does not establish exact fidelity.
Separate capabilities that support exploration from requirements that demand precision. A final logo or headline may be better added using controlled design tools.
Visionary Studios should evaluate the complete production workflow, including the parts completed outside the generator.
06 | WHAT DOES SCALABILITY MEAN?
Scalability is the ability to support increased demand with suitable resources. Demand can grow through more users, more images, larger outputs or tighter deadlines.
Ask what happens to completion times, failure rates and costs as the workload grows. A system that works well for one designer may form long queues when ten designers submit requests.
Scalability does not mean unlimited capacity or zero additional cost. State the demand range being tested and the conditions under which it remains acceptable.
07 | IDENTIFY THE UNIT OF WORK
A request might produce one image or a batch of candidates. Different resolutions and sampling settings can take different amounts of work.
To compare capacity meaningfully, define the workload: for example, requests with the same output dimensions and candidate count under recorded model settings.
Counting requests alone can mislead if one request asks for a small draft and another asks for many large outputs. Include acceptance and editing when the goal is usable creative work.
08 | LATENCY, THROUGHPUT AND CONCURRENT DEMAND
Latency is the completion time for a request. Throughput is completed work per unit time. Concurrent demand means overlapping requests from multiple users or jobs.
A shared system may increase throughput by batching work, while individual users still wait longer. A fast isolated response is weak evidence about busy-period performance.
Record a range of waiting times rather than only an average. An acceptable average can hide occasional delays that disrupt a time-sensitive review meeting.
09 | A SIMPLE CAPACITY ESTIMATE
Assume one worker processes one standard request every 30 seconds, sequentially and without downtime. Its ideal capacity is 60 ÷ 30 = 2 requests per minute .
Three independent identical workers would have an ideal combined capacity of 6 requests per minute. This assumes no shared bottleneck and a workload that can be distributed.
If sustained arrival demand is 8 requests per minute, it exceeds that capacity by 2 per minute. The backlog would tend to grow under these simplified assumptions. These are classroom figures, not a prediction for a real service.
10 | CAPACITY NEEDS HEADROOM
Even when average demand is below nominal capacity, bursts and variable service times can create queues. Planning at the theoretical maximum leaves little room for variation or failures.
Headroom is spare capacity intended to absorb changes. The amount needed depends on the workload and response-time requirements.
Do not infer a guaranteed waiting time from a simple capacity ratio. Real queueing depends on how requests arrive, how long they take and how scheduling works.
11 | ADDING RESOURCES DOES NOT REMOVE EVERY LIMIT
Scaling up means increasing the resources of a worker, such as using a device with more suitable capacity. Scaling out means adding workers and distributing work.
Both can help particular workloads, but shared storage, network limits, licensing or service quotas can restrict the benefit. The work must actually be distributable.
More workers do not automatically shorten the sequential denoising path of one image. They may mainly allow more independent requests to be handled at once.
12 | RELIABILITY AND COST STILL MATTER
A scalable service should be evaluated for failures, recovery and predictable operation as well as throughput. A failed job may require a retry and more waiting.
Track resource use and cost under realistic demand. Costs can grow with usage, reserved capacity or operational support depending on the arrangement.
The booklet supplies no service prices or quotas. Identify the information the studio should obtain rather than inventing current commercial figures.
13 | WHAT ARE BRAND REQUIREMENTS?
A brand is represented through recurring visual and communicative choices. Requirements may include palette, typography, logo treatment, tone and a recognisable artistic direction.
For a campaign, a character or motif may also need to remain consistent. The case study explicitly connects consistency with brand identity and artistic coherence.
Not every requirement should be left to the generator. The studio can combine generated imagery with approved design assets and controlled layout processes.
14 | TURN A STYLE GUIDE INTO OBSERVABLE CHECKS
| Requirement | Evidence | Possible workflow |
|---|---|---|
| Coherent illustration treatment | Line, shading and texture follow agreed descriptors | Generation guidance and visual review |
| Recurring character remains recognisable | Stable identity features across varied scenes | Approved references and campaign review |
| Exact logo and headline | Approved artwork and wording are preserved | Add controlled assets during layout |
| Suitable brand palette | Final colours meet the defined requirement | Reference guidance, colour checks and editing |
These checks are teaching examples. A colour that looks approximately right in a preview is not proof of exact delivery values.
15 | FLEXIBILITY CAN CONFLICT WITH CONSISTENCY
Open exploration invites variation. A brand campaign constrains some of that variation to keep a coherent identity.
A system may need separate workflows: exploratory drafts with broad options and approved production settings with tighter references and review. This is a possible organisational choice, not a guarantee built into the model.
When making a trade-off, identify what is allowed to change and what must remain stable. More creative variation is not automatically better for every stage.
16 | BUILD A TEST MATRIX
Combine task coverage and workload conditions. Test several representative briefs at low demand, normal demand and a realistic busy period.
For each condition, record acceptance, brand failures, completion times, errors and review effort. Keep model versions and relevant settings documented.
This reveals interactions: a workflow might preserve quality but become too slow under load, or be fast at scale while producing too much work that fails the brand brief.
17 | APPLY IT TO VISIONARY STUDIOS
A recommendation should connect capabilities, capacity and creative requirements, with clear limits on what the trial establishes.
18 | RECOMMEND A WORKFLOW
In your book, propose a workflow for the fictional explorer campaign. Identify one flexibility requirement, one scalability test and three brand checks.
Use the capacity example to explain why adding workers may improve throughput without reducing one image’s denoising time.
Reveal a possible response
Support text-guided exploration and compatible reference guidance for recurring characters. Test representative concurrent requests while recording queues, completion times and failures. Review character identity, illustration treatment and final approved logo and text.
Three ideal workers can complete six standard requests per minute under the example assumptions. Independent jobs can be distributed, but each image may still require its own sequential sampling process.
Read alongside flexibility, scalability and consistency in the official 2027 booklet. Workload figures and the explorer campaign are independent teaching examples.
CHECK YOUR UNDERSTANDING | CAPABILITY, CAPACITY OR BRAND?