Glossary Trainer

SL and HL (with a separate HL-only set). The booklet prints a terminology list with no definitions. Defining each term in your own words, and applying it to the scenario, is part of the assessed work.

The glossary is really a revision checklist in disguise. The exam rewards accurate terminology used throughout your answers, and it rejects definitions copied from the booklet. So the goal is not to memorise a wording, it is to be able to define each term in your own words and connect it to Visionary Studios. Use the tables below to check yourself, then close the page and write your own version.

Table of Contents

  1. How to learn a term properly
  2. SL + HL terms
  3. HL Section: HL-only terms
  4. Name the term: SL + HL
  5. Name the term: HL only (HL)
  6. Quick check
  7. Practice exercises
    1. Core
    2. Extension
    3. Challenge
  8. Connections

How to learn a term properly

For each term, do not stop at a one-line definition. The summer-task routine works well: for every term, note its definition, its purpose, its key characteristics, its advantages and disadvantages, a real-world application, and a generative-AI example tied to the scenario. A term you can use in a sentence about Visionary Studios is a term you own.

Two ways to lose terminology marks: copying the booklet’s wording when asked for a definition, and making up terms that do not exist. Write your own definitions, and only use terms you can actually explain.

SL + HL terms

Every student needs these. The “scenario link” column is a prompt, not a full answer; expand each one yourself.

Term In plain English Scenario link
Diffusion model Generates an image by starting from noise and removing it step by step The studio’s stable, high-quality default
Denoising Gradually turning a noisy image into a clear one The core repeated step of a diffusion model
Noise injection Adding random noise as the starting point for generation Gives each generated image a different seed
DDPM The framework for adding noise in training and removing it in generation What modern diffusion tools rely on internally
Convolutional neural network (CNN) An image-focused network that detects features with sliding filters The neural denoiser inside the diffusion model
Text-to-image generation Creating an image from a written prompt Turning a brief into draft concepts quickly
Conditional image generation Generating from a specific input such as a label or sketch Producing predictable, on-brief results
Unconditional image generation Generating with no steering input, from learned patterns Free exploration and synthetic datasets
Image-to-image translation Transforming one image into another while keeping its structure A sketch becomes a finished product shot
Segmentation map A labelled layout showing what belongs where Placing a product and logo precisely in a scene
Dataset curation Selecting and cleaning training data so it is legal and representative Avoiding copyrighted or biased training images
Bias mitigation Steps to reduce unfair or skewed outputs Making sure generated people are not stereotyped
Character consistency Keeping a recurring character looking the same across images A mascot identical across a campaign
Embedding-based approach Capturing an identity or feature as a learned vector for reuse The handle that keeps the mascot consistent
Training stability How reliably a model trains without breaking down A reason to prefer diffusion over a GAN

HL Section: HL-only terms

HL Only. These terms cover GANs and hybrid models. SL students are not expected to know them and can skip this section.

Term In plain English Scenario link
Generative adversarial network (GAN) Two networks trained against each other to make sharp images An option where sharpness matters most
Generator The network that creates images from a noise vector Produces the candidate images
Discriminator The network that judges real versus fake Pushes the generator to improve
Adversarial dynamic The competitive push-and-pull that drives a GAN What makes GANs powerful but unstable
D-dimensional noise vector The random vector a generator turns into an image A new vector gives a new image
Hybrid model A model combining approaches to get the benefits of each Quality plus control, at higher complexity
Variational autoencoder (VAE) A model that uses an organised latent space to generate and edit Smooth blending between design variations
Flow-based model A model with an exact, reversible noise-to-image mapping Precision and reproducibility when needed
Latent space A compressed, organised representation where similar images sit close Moving through it blends styles smoothly
Mode collapse A GAN failure where output variety collapses to a few images A risk to manage when training a GAN

Name the term: SL + HL

Read each description and type the term it defines. Use lowercase.

Type the term each description defines.

// Adding random noise as the starting point for generating an image
// Term: 

// Gradually turning a noisy image into a clear one
// Term: 

// Keeping a recurring character looking the same across many images
// Term: 

// Transforming a sketch into a realistic render while keeping its structure
// Term: 

// Selecting and cleaning training data so it is legal and representative
// Term:

Name the term: HL only (HL)

HL Only. This exercise uses HL terms.

Type the term each description defines. Use lowercase. (HL)

// The competitive push-and-pull between a generator and a discriminator
// Term: 

// A GAN failure where the generator produces only a few similar outputs
// Term: 

// A compressed, organised representation where similar images sit close together
// Term: 

// The network that judges whether an image is real or fake
// Term: 

// A model that combines several approaches to get the benefits of each
// Term:

Quick check

Q1. Which term means adding random noise as the starting point for generation?

Q2. Which approach most directly supports character consistency?

Q3. What should you do when a question asks you to define a glossary term?

Q4. (HL) Which term names a GAN failure where output variety collapses to a few images?

Practice exercises

Defining and applying terms is directly assessed. Write in your own words every time.

Core

  1. Define (4 marks) - In your own words, define: (a) diffusion model, (b) dataset curation. Then add one sentence applying each to Visionary Studios.
  2. State (2 marks) - State what a segmentation map is and one use for it.

Extension

  1. Distinguish (4 marks) - Distinguish between conditional and unconditional image generation, using the correct terms.
  2. Describe (3 marks) - Describe, in your own words, what training stability means and why it matters to the studio. Write in prose.

Challenge

  1. Explain (6 marks) (HL) - Using the HL terms accurately, explain how a generator, a discriminator, and the adversarial dynamic work together in a GAN, and where mode collapse fits in.

Connections

© EduCS.me — A resource hub for Computer Science education

This site uses Just the Docs, a documentation theme for Jekyll.