【Video】🎞️ 02. Why I Tested AI Video as a Teaching Medium — and Chose to Walk Away

topics: [“ai”, “education”, “design”, “prompt”, “runway”]

🧾 Conclusion First

AI-generated video is not suitable as the main medium for educational or technical materials.

This is not an impression or a gut feeling.
It is a design decision made after actually using the tools,
designing prompts, generating outputs, storing results,
and evaluating them from an educational standpoint.

🔍 Why I Decided to Test It

Generative AI is already practical in areas such as:

• Text generation
• Diagram assistance
• Code skeleton generation

in education and technical fields.

So the natural next question was:

Can AI-generated video be used for teaching?

If video could be effective for lecture introductions,
technical presentations, or research explanations,
it would be worth adopting.

To answer this, I created a dedicated verification repository
and tested it in practice.

🧪 What I Actually Did

The verification followed these principles:

• Prompts managed in Markdown
• Reproducibility as the top priority
• Comparison of Image → Video and Text → Video
• Generated outputs stored on GitHub
• Evaluation based on “usable or dangerous” for educational use

The key question was not
“Can it generate something?”
but
“Does it remain as a design asset?”

🧠 What I Learned Technically

① Output Is Not Stable

• Results fluctuate even with identical prompts
• Not suitable for comparison or verification

② Accuracy Cannot Be Guaranteed

• Text, symbols, and structures easily break
• “Looking right” overrides being correct

③ Credit-Based Systems Conflict with Verification

• Failures still consume credits
• Trial-and-error is psychologically discouraged

🚫 The Critical Problem for Educational Use

In educational and technical materials, the top priority is:

Do not mislead.

AI video excels at:

• Atmosphere
• Immersion
• Visual impact

But it is not a medium for accurately conveying:

• Mathematical expressions
• Structural relationships
• Causal chains

The conclusion was clear:

AI video can never be more than a “preface,”
not an explanation.

🧭 Final Policy Decision

Restrict AI Video to “Meaning-Free” Use Cases

• Lecture openings
• Presentation introductions
• Section transitions

The safety condition is simple:
do not put information into the video.

❌ What I Decided Not to Use It For

• Video explanations of equations
• Animated representations of technical structures
• Any explanation requiring accuracy

When in doubt,
do not use it.

🧱 Why Write About “Walking Away”

Discussions around generative AI tend to focus on:

• What was achieved
• How impressive it looks

But in real design work:

The reasons for deciding not to use something
are often the most valuable assets.

This verification clarified:

• The limitations of AI video
• The boundaries required in educational design
• Criteria for choosing the right battlefield

📂 Verification Logs and Generated Evidence

This article is based on actual repositories and generated videos.

Verification Repository

• GitHub: AI Video Lab
  👉 https://github.com/samizo-aitl/ai-video-lab

This repository includes:

• Reproducible prompt designs (prompts/)
• Generation workflows (workflows/)
• Failure cases and evaluation notes (notes/)
• Generated samples (samples/)

Generated Sample Video

• Reference video (MP4, 5s, Image → Video)
  👉 https://github.com/samizo-aitl/ai-video-lab/blob/main/samples/output/abstract_lab_01.mp4

• Playable directly on GitHub
• An intentionally meaningless abstract video for educational introductions

🧾 Summary

• AI video can be generated
• But it should not be the core of educational material
• Choosing not to use it is not a failure
• It is a design outcome

What matters in the age of generative AI is not:

what we can use

but rather:

where we draw the line, and how clearly we can explain it.