Provider-independent production workflow for AI agents creating educational animated lessons, classroom explainers, STEM visualizations, history/social-science animations, training modules, microlearning clips, whiteboard-style explainers, diagram-driven videos, and generated or assembled instructional animation assets. Use when planning, scripting, storyboarding, generating, reviewing, localizing, or QAing animation whose primary purpose is learning.
10k tokens
context cost
the whole folder, loaded on every use
2
files
instructions only
0
copies elsewhere
how many repositories repackaged it
112
stars on the repo
on the repository, not the skill itself
Install
one command, takes just this skill from the repository
Use this skill to make animated instruction accurate, learnable, accessible, and production-ready. Treat the animation as an instructional product first and a media product second: every visual, motion cue, narration line, and assessment moment must support a defined learning objective.
This is provider-independent. Apply it whether the final piece is built with generated video, generated images plus motion design, vector/SVG animation, whiteboard animation, slides-to-video, Manim/D3/STEM visualization, character animation, screen capture, or a compositing runtime.
First decisions
Before writing prompts, scenes, or code, establish the instructional contract:
Define the learner, not just the topic.
Age/grade or job role.
Prior knowledge and prerequisites.
Language proficiency and reading level.
Accessibility needs and likely viewing context.
Motivation: why they need this now.
State 1-3 measurable learning objectives.
Use observable verbs: identify, compare, solve, explain, predict, classify, apply.
Avoid vague objectives such as "understand gravity" unless converted into a visible performance: "predict how changing mass affects gravitational force in a two-body diagram."
Decide the evidence standard.
Classroom/STEM/history/training: cite authoritative sources for claims.
Medical, legal, financial, safety, regulated training, product efficacy, or compliance topics: require subject-matter expert review before final delivery.
Commercial or product-training claims: substantiate objective claims before publication.
Compare alternatives: before/after, misconception/correction, model A/model B.
Guide attention through a diagram or formula.
Demonstrate steps: worked example, procedure, tool flow.
Make practice possible: pause, question, feedback, next step.
If the topic is emotionally sensitive, politically contested, identity-related, high-stakes, or likely to be oversimplified, add an explicit review gate with an SME, educator, compliance reviewer, or community reviewer before asset generation.
Learning design principles
Ground the production in these documented principles, then adapt with professional judgment.
Documented facts and research-backed guidance:
Learners bring prior knowledge; effective instruction surfaces it and builds from it. Use a quick diagnostic prompt, misconception check, or familiar analogy before the new model.
Cognitive load rises when learners must process irrelevant information or confusing layout. Reduce decorative motion, redundant on-screen text, and scene clutter when the learner is processing a new concept.
Multimedia learning research supports coherence, signaling, spatial contiguity, and temporal contiguity: remove extraneous material, cue what matters, place labels near the thing labeled, and time narration with the visual event it describes.
Universal Design for Learning encourages multiple means of engagement, representation, and action/expression. In animation practice, this means offer more than one way to access the idea: narration, captions/transcript, diagram, examples, and practice.
Production heuristics:
Prefer one main idea per scene. If a scene needs more than one sentence to explain its purpose, split it.
Use motion to show change, causality, sequence, or attention. Do not animate everything because animation is available.
Build from concrete to abstract: familiar scenario -> visual model -> formal term/formula -> transfer example.
Repeat the learning objective at the opening, midpoint, and closing in different forms.
Treat labels, arrows, colors, and motion paths as instructional language, not decoration.
Script and sequence pattern
A strong animated lesson usually follows this sequence. Modify it when the domain demands another structure.
Hook with relevance and a question.
Name the learner's problem: "Why does your phone battery drain faster in the cold?"
Ask a question the lesson will answer.
Activate prior knowledge.
Use one familiar object, prior lesson link, or misconception.
Do not shame misconceptions; frame them as useful starting models.
State the learning objective.
"By the end, you will be able to..."
Introduce vocabulary only when needed.
Pair each term with a visual referent and one example.
Animate the core mechanism.
Show only the necessary parts first.
Add labels and variables as they become relevant.
Pause for a check for understanding.
Ask the learner to predict, classify, choose the next step, or explain why.
Provide the answer or feedback after a beat.
Apply to a new case.
Transfer is the point: show the idea outside the opening example.
Summarize and give a next action.
Restate the key mental model.
Invite a practice problem, reflection, or next lesson.
For microlearning under 90 seconds, keep one objective, one core model, one check for understanding, and one transfer example. For longer training modules, create chapters and insert practice every 2-4 minutes.
Storyboard and visual grammar
For each scene, specify:
objective served;
narration line or learner-facing text;
visual model;
motion event;
labels/formulas;
accessibility notes;
source/citation needs;
assessment moment, if any;
asset prompts or build instructions.
Use this visual grammar:
Establish the whole before zooming into parts.
Keep spatial consistency. If "input" starts on the left and "output" on the right, keep that mapping unless the lesson explicitly changes frames of reference.
Reserve color for meaning. Example: blue = known quantities, orange = unknown, green = verified answer. Never rely on color alone.
Place labels next to the object or process they name. Avoid legends that force the learner to look back and forth.
Reveal formulas progressively:
show the physical situation;
label quantities;
introduce the equation;
substitute numbers;
animate simplification;
interpret the result in words.
In history/social-science animation, separate chronology, causality, and perspective. Use timelines for sequence, maps for geography, quotation/source cards for evidence, and explicit language for uncertainty or contested interpretation.
In procedural training, show the environment, the action, the decision rule, and the consequence. Do not only show the "happy path"; include common errors when safety or competence depends on avoiding them.
Narration, pacing, and text
Write narration for the ear:
Use short spoken sentences.
Put the main message early.
Define terms before using them in compound explanations.
Read the script aloud and time it. Dense scientific narration often needs slower pacing than ordinary voiceover.
Pause before and after new diagrams, formulas, key terms, and questions.
Use on-screen text sparingly:
Captions are not a replacement for thoughtful visual labels.
Do not put full narration paragraphs on screen while the narrator says the same thing, unless the audience specifically needs verbatim reinforcement.
Put only keywords, variables, labels, short steps, or question prompts on the canvas.
Keep text large enough for the target platform and device.
Audio mix:
Keep narration intelligible over music and effects.
For speech-first educational media, background audio should be absent, user-controllable, or clearly lower than narration.
Use sound effects only when they reinforce meaning, such as a click when a variable locks into a formula or a gentle chime before a question.
Diagrams, formulas, data, and labels
Make every diagram defensible:
Check the domain model before designing the visual metaphor.
Mark simplifications: "not to scale," "simplified model," "one possible pathway," or "approximate."
Use units consistently.
Avoid false precision in numbers or charts.
Make axes, legends, and scales readable.
Use direct labels, not unexplained color coding.
For generated assets, never trust rendered text, formulas, maps, flags, chemical structures, anatomical details, historical artifacts, or charts without manual verification.
For STEM animation:
Use actual equations and units in a text layer or deterministic render whenever possible. Do not rely on image/video models to draw formulas.
For geometry, graphs, maps, code, or data visualizations, prefer deterministic tools (SVG, D3, Manim, plotting libraries, GIS, composition code) over free-form image generation.
When animating a process, ensure conservation laws, arrows, causal direction, and scale relationships are accurate.
For history/social science:
Cite primary sources or reputable scholarly/official summaries.
Distinguish event, interpretation, and disputed claim.
Avoid flattening groups into stereotypes or single motives.
Include perspective markers when relevant: "from the government's view," "as reported by," "historians debate."
Generated-asset prompt translation
Convert instructional intent into asset prompts without losing accuracy.
Use a three-layer prompt:
Instructional function: what the asset teaches.
Visual specification: what should appear and how it is composed.
Accuracy constraints: what must be correct or avoided.
Example structure:
Instructional function: Show that electrical current is charge flow through a closed circuit, not "used up" by the bulb.
Visual specification: Clean flat vector animation frame, battery on left, switch at top, bulb on right, wire loop, small blue electrons moving around the whole loop, warm glow at bulb.
Accuracy constraints: Do not show electrons disappearing in the bulb. Do not show current leaking into air. Include labels: battery, switch, bulb, closed loop. Use separate editable text layers if possible.
When using generative image/video tools:
Ask for clean negative space where labels will be added later.
Keep generated text out of the prompt unless the model reliably supports typography; add text in post.
Request simple shapes for diagrams and complex aesthetics for mood only when mood does not carry factual load.
Generate diagrams in parts if the model struggles: background, objects, arrows, labels, overlays.
Preserve seeds/model IDs/provider settings and record asset provenance.
Review every generated asset against the learning objective and factual source before it enters the edit.
Accessibility and inclusion requirements
Plan accessibility from the script stage, not after render.
Minimum checks:
Captions: provide synchronized captions for spoken audio and meaningful non-speech sounds.
Transcript: provide a transcript for learner review and search.
Audio description or integrated description: if essential information is visual-only, describe it in narration or provide an audio-described version.
Flashing risk: avoid flashes and rapid high-contrast flicker. If unavoidable, test against WCAG flashing thresholds and add a warning only as a last resort; warning does not replace risk reduction.
Color and sensory independence: do not rely only on color, shape, position, or sound to convey instructions.
Contrast and legibility: check text and diagrams at the final delivery size.
Timing: give learners enough time to read labels, inspect diagrams, and answer questions.
Motion comfort: avoid unnecessary camera shake, fast zooms, and continuous parallax; provide a reduced-motion variant when the platform or product supports it.
Language: use plain language for the intended audience and define technical terms.
Inclusive representation:
Show learners, workers, scientists, historical actors, and communities with dignity and specificity.
Avoid tokenism: representation should fit the setting, not appear as decorative diversity.
Check names, clothing, artifacts, maps, religions, cultural practices, disability representation, and historical context for accuracy.
Avoid using accent, dialect, costume, body shape, skin tone, or disability as a joke or shorthand for ability.
For global audiences, avoid culture-bound idioms, sports metaphors, gestures, or food references unless explained or localized.
Assessment and feedback moments
Educational animation should not only tell; it should let the learner test a mental model.
Use checks for understanding that match the objective:
Identify: "Which part stores the energy?"
Predict: "What happens if the switch opens?"
Compare: "Which graph shows constant speed?"
Explain: "Why did the denominator increase?"
Apply: "Try the same rule on this new example."
Detect misconception: "What is wrong with this diagram?"
Give feedback:
Confirm the correct answer.
Explain the reason, not just the option letter.
For incorrect options, name the misconception gently.
In asynchronous video, pause with "Pause now and choose" before revealing the answer.
In interactive delivery, record expected answer states and remediation branches.
Localization and delivery variants
Plan variants before final layout:
Translation can expand, contract, or reflow text substantially depending on language; leave adaptable space and test layouts with the target locale.
Avoid baking text into raster images when localization is likely.
Keep captions, transcript, narration script, glossary, and on-screen text in separate editable files when possible.
Localize examples, units, currency, dates, names, idioms, maps, cultural references, and legal/regulatory terms.
Re-check claims and compliance rules for each region and platform at production time.
For right-to-left languages, plan mirrored layouts deliberately; do not mirror scientific diagrams, maps, formulas, or UI screenshots unless the domain requires it.
For sign language or audio-described variants, reserve frame space and timing.
Delivery variants to consider:
classroom version with pauses and discussion prompts;
self-paced version with more feedback;
social/microlearning version with one objective and burned-in captions;
LMS version with quiz metadata and transcript;
accessibility variant with extended audio description or reduced motion;
localization package with script, captions, on-screen text, glossary, sources, and editable project files.
Review gates and escalation
Require review before final delivery when:
The lesson teaches medical, legal, financial, safety, engineering, compliance, or certification content.
It makes commercial, efficacy, performance, health, or comparative claims.
It depicts protected classes, cultural practices, historical trauma, political conflict, religion, colonialism, war, or current events.
It uses copyrighted, trademarked, celebrity, student, patient, employee, or client-provided material.
It uses AI-generated content where authorship, licensing, consent, or disclosure requirements affect publication.
The production will be used for graded instruction, professional accreditation, public policy, workplace discipline, or child-directed learning.
Escalate with a concise package:
learning objectives;
script;
storyboard;
source list;
claim table;
generated-asset provenance;
accessibility plan;
unresolved questions.
Do not present legal, medical, financial, or regulatory conclusions as advice. Ask for qualified review.
Educational animation QA
Run this checklist before final render and again after export.
Instructional QA:
Each scene maps to a learning objective or is removed.
The opening says what the learner will be able to do.
Prior knowledge or misconception is addressed.
New terms are defined before they are needed.
The sequence moves from concrete to abstract or otherwise justifies its order.
At least one check for understanding appears in each substantial lesson.
Feedback explains why the answer is correct.
The summary supports transfer to a new case.
Factual QA:
Claims are sourced and up to date.
Numbers, dates, formulas, units, maps, diagrams, names, and terminology are verified.
Simplifications are labeled.
Generated text and diagrams are manually checked.
SME/compliance review is complete when required.
Animation QA:
Motion clarifies cause, sequence, comparison, or attention.
Labels appear near the visual element they identify.
Narration and animation are temporally aligned.
The learner has time to inspect each diagram.
The edit does not introduce contradictions with the script.
Visual metaphors do not imply false facts.
Accessibility QA:
Captions are synchronized, punctuated, speaker-labeled where needed, and include meaningful sounds.
Transcript is complete.
Audio description or integrated narration covers essential visual-only information.
Contrast, text size, and final-device readability pass.
No instruction depends only on color, shape, sound, or position.
Flashing/flicker risk is removed or tested.
Audio mix keeps speech clear.
Reduced-motion or alternate access is provided when needed.
Rights/provenance QA:
Sources and citations are stored with the project.
AI providers, model names, prompts, seeds/settings, and generation dates are recorded.
Third-party assets have licenses and attribution requirements documented.
Releases/permissions are present for people, voices, private locations, or client assets.
Copyrightability, disclosure, platform, and institutional policy questions are escalated rather than assumed.
Example: STEM microlearning clip
Production intent: a 75-second animation for middle-school learners explaining why seasons are caused by Earth's axial tilt, not distance from the Sun.
Inputs and constraints:
One objective: learners can identify axial tilt as the cause of seasons.
Avoid common misconception: "summer happens because Earth is closer to the Sun."
Delivery: classroom and short-form video.
Workflow:
Open with a misconception check: "If Earth is closest to the Sun in January, why is it winter in the Northern Hemisphere?"
Show a simple Sun-Earth diagram with Earth's orbit nearly circular and axial tilt held constant.
Animate sunlight hitting the Northern Hemisphere more directly during June and less directly during December.
Add labels only after the visual relationship appears: axial tilt, direct light, spread-out light.
Ask: "Which hemisphere gets more direct light now?"
Reveal answer and apply to the Southern Hemisphere.
End with: "Seasons come from tilt changing the angle and duration of sunlight."
Generated-asset prompt:
Create a clean educational vector-style space diagram for an animation about seasons. Show the Sun on the left and Earth in four orbital positions around it, with Earth's axis tilted the same direction at every position. Leave empty margins for labels. Use simple geometry, not photorealism. Do not exaggerate the orbit into a long oval. Do not show Earth getting dramatically closer to the Sun. No embedded text.
Post-production notes:
Add labels and arrows in the compositor, not the image generator.
Verify orbit shape, tilt direction, and hemisphere labels.
Provide captions, transcript, and integrated description such as "Notice the axis keeps pointing the same way as Earth moves around the Sun."
Likely failure modes:
Generated image makes the orbit too elliptical.
Earth axis points different directions at different orbit positions.
Narration says "more sunlight" without specifying angle and duration.
Example: workplace training module
Production intent: a 4-minute animated training module on safe ladder setup for new warehouse employees.
Inputs and constraints:
Requires safety review by the client or qualified safety SME.
Objective: learner can choose a safe ladder angle, inspect the ladder, and identify two unsafe setups.
Delivery: LMS with captions, transcript, and quiz.
Storyboard excerpt:
| Scene | Learning function | Visual | Assessment |
|---|---|---|---|
| 1 | Relevance | Worker pauses before using ladder; narrator says falls are preventable when setup is checked | None |
| 2 | Inspection | Close-up of feet, rungs, labels, surface | "Which item means stop and report?" |
| 3 | Angle | Side diagram showing ladder, wall, floor, angle cue | "Which setup is safest?" |
| 4 | Common error | Ladder on box, wet floor, blocked doorway | "Name two hazards." |
| 5 | Transfer | New warehouse aisle with similar rule | "What would you check first?" |
Asset prompt for a hazard-identification scene:
Instructional function: Show an intentionally unsafe ladder setup for a warehouse safety lesson.
Visual specification: Flat 2D training animation style, warehouse aisle, extension ladder leaning against a shelf, wet floor sign nearby, small box placed under one ladder foot, worker standing back and evaluating, no injury shown.
Accuracy constraints: Make hazards visible but not sensational. Do not show anyone climbing the unsafe ladder. Leave space for callout labels. No embedded text.
Review package:
Source or client procedure for ladder rules.
Claim table for every safety instruction.
SME sign-off before final render.
Accessibility pass for captions, transcript, audio description, and flashing risk.
Likely failure modes:
Animation accidentally models unsafe behavior as if it were acceptable.
The assessment asks recall but the objective requires hazard recognition.
Background music masks instructions.
Example: history explainer
Production intent: a 3-minute classroom animation explaining how one historical policy affected multiple groups differently.
Inputs and constraints:
Topic has contested interpretations and identity implications.
Objective: learner can distinguish event chronology from historical interpretation.
Requires sources and educator review.
Approach:
Start with a timeline of dated events.
Use maps only where geography matters; label borders and note if they are modern approximations.
Introduce primary-source snippets as evidence cards, with short paraphrase and citation.
Use multiple perspective cards: government officials, affected community, later historians.
Mark uncertainty: "Records from this group are limited," or "Historians disagree about..."
Ask learners to classify statements as event, cause, consequence, or interpretation.
Prompt for a non-factual mood background:
Create a subdued classroom-friendly animated background: paper texture, archival-inspired neutral palette, slow parallax of abstract document shapes and map-grid lines. No readable text, no real flags, no portraits, no symbols associated with a specific group. Leave center area clean for verified timeline and source cards added in post.
Why this works:
The generated asset carries atmosphere but not factual content.
Dates, maps, quotations, names, and claims are added as verified editable layers.
The assessment checks historical reasoning rather than memorization only.
Source and verification notes
Sources consulted and verified on 2026-07-11:
W3C Web Content Accessibility Guidelines (WCAG) 2.2: https://www.w3.org/TR/WCAG22/
W3C Making Audio and Video Media Accessible: https://www.w3.org/WAI/media/av/
W3C Media Accessibility User Requirements: https://www.w3.org/TR/media-accessibility-reqs/
CDC Clear Communication Index User Guide: https://www.cdc.gov/ccindex/pdf/clear-communication-user-guide.pdf
Digital.gov Plain Language guides: https://digital.gov/guides/plain-language
CAST Universal Design for Learning Guidelines 3.0: https://udlguidelines.cast.org/
National Academies, How People Learn: https://www.nationalacademies.org/read/9853
Sweller, van Merrienboer, and Paas, "Cognitive Architecture and Instructional Design: 20 Years Later": https://link.springer.com/article/10.1007/s10648-019-09465-5
Cambridge Handbook of Multimedia Learning, chapter on coherence, signaling, redundancy, spatial contiguity, and temporal contiguity: https://www.cambridge.org/core/books/abs/cambridge-handbook-of-multimedia-learning/principles-for-reducing-extraneous-processing-in-multimedia-learning-coherence-signaling-redundancy-spatial-contiguity-and-temporal-contiguity-principles/CD5B7AE1279A9AB81F8EEBB53DBEC86E
FTC Advertising and Marketing guidance: https://www.ftc.gov/business-guidance/advertising-marketing