How to Create Educational Videos for YouTube (Script, Visuals, and the Production Loop)

Most educational videos are not badly made. They are badly paced.

The information is right, the slides are tidy, and somewhere around the fifty second mark the viewer realizes they are being read a document. They leave, YouTube notices, and the video that took a weekend reaches four hundred people.

A teaching video has one job: keep a person watching long enough to understand the thing. Everything below is in service of that.

Here is a short one we built while writing this guide. It answers a question people actually type into a search box, in under a minute, with nothing on screen but drawings.

Start from the question, not the topic

"Databases" is a topic. "Which database should I use?" is a question, and it is the one somebody typed.

Write the question down first, in the words a learner would use, before you write anything else. It does three things at once: it becomes the title, it becomes the first line of narration, and it becomes the test for every scene you are about to make. If a scene does not move the viewer closer to the answer, it is decoration.

The video above opens on the question and answers it in the same breath: it depends on the shape of your data. The viewer now knows what they are going to get and roughly how long it will take. That is the contract, and the first ten seconds are where you sign it.

Write the script before anything moves

Animating an unfinished script is the most expensive mistake in this format, because every rewrite invalidates a visual.

Write the whole thing as spoken sentences, out loud, in one pass. Then cut it by a third. Educational writing bloats in predictable ways: throat-clearing intros, hedges ("it's worth noting that"), and the same idea stated twice for safety. All three cost you retention.

A structure that reliably works for one concept:

1. The question, in the viewer's words.
2. The one line answer, immediately. Do not save it for the end.
3. One chapter per sub-idea, in the order a beginner needs them, not the order an expert would list them.
4. A comparison or a decision table, so the viewer leaves with something they can act on.
5. The recap, which is the same answer as step two, now with meaning behind it.

That is the shape of the database explainer: a question, the answer, six chapters of one shape each, and a picker at the end that maps a situation to a choice. If you have written a product launch video before, you will recognize the discipline. The beats are different but the ruthlessness is the same.

Draw the thing. Do not list it.

The difference between a lesson and a slide deck is whether the visual carries the idea or repeats the words.

A bullet that says "documents can nest data" is the narration in text form. A drawing of a JSON object with an items array inside it, appearing as the sentence is spoken, is the idea itself. The viewer understands nesting because they saw something nested.

Some rules that hold up across subjects:

• One visual per idea, and it stays until the idea is finished. Swapping the picture mid sentence costs comprehension.
• Build it in front of them. A diagram that draws itself as it is described is followed far more easily than one that appears complete.
• Label sparingly and largely. If the text is too small to read on a phone, it is not on screen, it is just noise.
• Use one accent color for meaning. In the video above, red means "the thing to notice": the foreign key, the query, the latency. Everything else is ink on paper.
• Show the comparison at the end. Learners remember the table.

Hand drawn is a style choice, not a requirement, but it earns its keep here. A sketch reads as thinking out loud, which is exactly the register a good explanation wants. Our guide to motion graphics without After Effects covers the design rules that separate a professional looking animation from an amateur one, and they apply to whiteboard style work just as much as to product marketing.

Let narration set the pace

In a teaching video the voice is the clock. The animation follows it, never the other way around.

Record the narration first, or generate it, then time the visuals to the words. When the line says "linked by keys", the arrow should be drawing. A diagram that lands two seconds late reads as a mistake even to a viewer who could not tell you why.

If you are generating the voice, spend time on the model rather than on the script hedges. The current options and what each costs are in our roundup of the best AI voice generation models, and the same guide's advice holds: pick a voice that sounds like a person explaining something, not a person announcing something.

Music is optional and should be quiet enough that you notice it only when it stops. The video above runs a bed at 22% volume under the narration. If you want to source one, we compared the tools in best AI music generation models.

Chapter it, because chapters are a writing test

YouTube chapters are a viewer convenience and a search surface, and they require the first timestamp to be 00:00, at least three timestamps in ascending order, and a minimum chapter length of ten seconds (YouTube Help).

The useful part is what writing them reveals. If a chapter needs two titles, it is two chapters. If two chapters want the same title, they are one. Chaptering a lesson before you animate it is the cheapest structural edit you will ever make.

Write the outline as timestamps even when the video is too short to publish them. The database explainer runs six chapters in 56 seconds, which is well under YouTube's ten second minimum per chapter, so nothing goes in the description. It still got written this way first:

Which database should you use?  1. Relational  2. Document  3. Key value  4. Graph  5. Time series  6. VectorHow to pick one

Expand the same outline into a ten minute lesson and every line becomes a real chapter.

Publish it properly

The video is most of the work. These are the parts that decide whether anyone sees it.

• Thumbnail. 1280x720, 16:9, JPG or PNG (YouTube Help). Put three or four words on it at most, at a size that survives a phone-sized card.
• Title. The question, phrased the way it is searched. "Types of databases explained" beats "A journey through data storage".
• Description. The one line answer in the first sentence, then the chapters, then the links. The first sentence is what shows in search.
• Captions. Upload an SRT rather than relying only on auto captions. It makes the lesson watchable with sound off and searchable inside YouTube.
• End screen. Point at the next lesson in the series, not at your homepage.

The production loop

This is where most people stall. The script is good, the structure is clear, and now it has to become moving pictures.

The loop we use is the one behind every video in our template gallery: describe the lesson to an agent, look at the frames it produces, and correct it in plain language until it is right.

In practice, for the database explainer:

• The project is eight scenes, one per chapter, on a 1920x1080 canvas at 30fps. Total runtime is 56 seconds.
• Each scene is real code you can open and read. The engine draws the sketch geometry, the timing lives in the project manifest, and the narration sits on its own audio track with the music underneath it.
• Corrections are sentences. "The foreign key arrow should land after the highlight, not before" is a change you ask for, not a keyframe you drag.
• The export runs on your own machine and is frame for frame identical to the preview, so what you approved is what uploads.

The whole layered picture of what goes into a finished AI video, the scene, the voice, the music, the sound effects, is in our guide to AI video generation. Teaching videos use fewer of those layers than a launch film does, which is part of why they are the best format to start with.

Start from a lesson that already works

The fastest way to make your first explainer is to take one apart.

The types of databases template is the full project: eight scenes, the narration track, the music bed, the sketch components. Remix it, replace the subject with yours, and the structure, pacing, and type are already decided. There are more in the educational templates collection.

If you want the studio itself, it runs on macOS on Apple silicon and uses your own Claude Code or Codex subscription. There is more about the workflow on the educational videos page, or you can download GenMotion and have your first lesson rendered tonight.