AI4K12 · Five Big Ideas in AI
Reference Models Stay In-Family
A reference model must be trained in the same family of AI tools used to make the images. It does not carry over to a different family.
Image Generation
Visual Language and Consistency
Overview
# Overview
Curriculum.UnitsSpecifying a film’s visual style in words and generating a consistent image sequence from it. Students define aesthetic, lighting, mood and palette, train a reference for character consistency, and control drift and cost across a twelve-image spine.
# Objectives
Curriculum.Units- 01 By the end, you can Define a film’s look in words — style, lighting, mood, palette, atmosphere — drawn from a real film and a named visual style.
- 02 By the end, you can Direct an AI image tool from that written style, using a trained reference so a character stays consistent across shots.
- 03 By the end, you can Detect and correct style or character drift, editing only what changed, and keep the set within its credit budget.
- 04 By the end, you can Combine both to generate the twelve-image spine that all reads as one film.
# Sessions
Curriculum.Sessions- How AI Images Are Made
- Quiz Real Or Fake Face Off Students look at pairs of faces on screen and vote on which one was made by an AI before the answer is revealed.
- Demo From Static To Picture The teacher shows a model turning a screen of random static into a clear picture step by step so the class sees how it builds an image from noise.
- Worksheet AI Tell Hunt Students move around a set of printed images and mark the giveaway mistakes like odd hands, garbled text, and warped backgrounds that hint a picture is AI made.
- Comparing Image and Video Models
- Demo Same Prompt Bake Off The teacher runs one identical prompt through two or three different models and the class compares how good, how fast, and how expensive each result is.
- Worksheet Cost Per Image Worksheet Students fill in a table of each model's credit price and work out the full cost of a batch before making anything.
- Debate Which Model For The Budget Two teams argue which model gives the best value for a fixed pile of credits and the class picks a winner.
- Writing the Visual Style
- Game Describe To Reproduce Students describe a picture in words carefully enough that an AI tool can recreate it closely.
- Demo Style Breakdown Demo The teacher walks the class through a full breakdown of one film's look as a model example.
- Keeping Characters Consistent
- Game Spot The Drift Teams make one character three times, spot what changed, then fix only that part in the editor.
- Building the Twelve Image Spine
- Build Spine Review The class lines up twelve images side by side and remakes any that do not match the film's look.
Applied · Film
# Knowledge & Skills
ACA-AppliedA style description names five things: the overall style, the lighting, the mood, the colour palette, and the atmosphere.
You can find words to describe a look in a published list of named visual styles, instead of inventing them.
The twelve-image backbone shows the whole world of the story, not only the main character.
The shape of the image (its aspect ratio, such as widescreen or square) stays the same across the whole set.
Keeping a character and world looking the same depends on noticing when something has drifted, not just knowing that it matters.
Access to the image tool and its credit cost start at the keyframe stage (the first image made for a shot), before any video clip is made.
Make the twelve images that form the backbone of the story. point w2d2-s2-p1; evidence w2d2-e1
Gather those words from a real film and a named visual style. point w2d2-s1-p2; evidence w2d2-e4
Five-Axis Style Definition Aesthetic, lighting, mood, colour palette, ambience — written once and shared.
Same-Family Training Rule A reference model must be trained in the same model family it will generate from; it doesn't transfer across families.
Twelve-Image World Spine The reference set covers the whole world of the story, not only the protagonist.
Write From the Shared Definition Every prompt draws from one team style definition, not invented per shot.
AI
# AI Concepts
ACA-AI.ConceptsAligned with AI4K12 · Five Big Ideas in AI View framework →
AI4K12 · Five Big Ideas in AI
Prompts From the Shared Style
Every prompt (the instruction you type into an AI tool) in this unit comes from the shared style description, not written on its own.
# AI Competencies
ACA-AI.CompetenciesAligned with UNESCO · AI Competency Framework for Students View framework →
AI Techniques and Applications
4.3.3 Creating AI Tools
Developing advanced technical creativity to design, build, and innovate with AI tools and methods.
# AI Skills
ACA-Tools.SkillsAligned with Derived from course activity
Derived from course activity
LoRA Training & Prompt-Driven Image Control
Training a LoRA matched to the target image-generation model, and diagnosing prompt failures when generated characters/scenes go wrong. Evidence: "you must generate a LoRA in the same model as youre going to use for image generation" and "we had problem here because of a bad prompt we did, and started making so many strange faces" (image-generation/animating-stills).
# Competency Goals
ACA-AI.GoalsAligned with UNESCO · AI Competency Framework for Students View framework →
UNESCO · AI Competency Framework for Students
CG4.3.3.3 Testing and Comparing AI Tools
Rigorously evaluating different AI approaches, tools, and implementations to determine the best solution for a problem.
UNESCO · AI Competency Framework for Students
CG4.3.3.1 Advanced Dataset and Technique Skills
Mastering sophisticated approaches to data preparation, feature engineering, and advanced AI algorithms and techniques.
UNESCO · AI Competency Framework for Students
Describe a Look in Words
Describe a look in words an AI tool can use: style, lighting, mood, colours, and atmosphere.
UNESCO · AI Competency Framework for Students
Train a Character LoRA
Train a small add-on (a LoRA) in the same AI tool so one character looks the same in every shot.
UNESCO · AI Competency Framework for Students
Price the Generation First
Check the credit cost (the pay-per-use amount an AI tool charges) of making an image before you spend it.
UNESCO · AI Competency Framework for Students
Spot Character Drift
Spot what changed between two versions of the same character before fixing it.
Deliverable
# Image Generation
Projects.BriefsProduce twelve images that look like they belong to the same film, forming the visual spine of the trailer. A set of twelve final-quality images, all in one fixed aspect ratio, covering the world of the film and the main character.
Transfers to: Naming a visual style precisely enough for a model to act on it.
# Assessed toward these aims
Projects.RubricsThis deliverable is where students demonstrate
- B Specify a visual style and generate a consistent image sequence from itLook out for
- Style consistency: All twelve images look like they belong to the same film. The style, lighting, mood, colour palette and atmosphere stay the same across the set, and the aspect ratio does not change.
- Style definition in words: The look is described across the five things (style, lighting, mood, colour palette, atmosphere) in words an AI tool can use, drawn from a real film and a named visual style category.
- Character consistency: The character looks the same across shots. A reference model is trained using the same family of AI tools used to generate the images.
- Drift control: Differences between two versions of a character are found and corrected before the work is handed in. Only the part that changed is edited, to match it back to the reference image.
- Cost control: The credit cost of making an image is checked before it is spent, and the set is delivered within budget.