I Finished a 3-Hour Livestream. AI Turned It Into a YouTube Video in 10 Minutes.
During the livestream, AI organized everything in real time. When the stream ended, AI4PKM and GOBI Desktop automatically generated a YouTube summary video.
AI Turned My 3-Hour Livestream Into a YouTube Video in Just 10 Minutes
Right after finishing a 3-hour livestream,
AI organized the entire conversation and produced a YouTube summary video in just 10 minutes.
Even when I talk freely and without much structure,
AI captures everything in real time.
It organizes the ideas,
summarizes them systematically,
visualizes the structure of the conversation,
and as soon as I stop speaking,
it generates a YouTube-ready video.
For a long time, this was something I had only imagined.
But last week,
it actually happened.
To me.
Here is the result:
This video was generated by AI about 10 minutes after my 3-hour YouTube livestream ended.
AI Is Fluent in Every Language
My livestreams are normally conducted in Korean.
The first step was to create a Korean summary video from the 3-hour livestream.
Once AI generates the video,
translating it into other languages becomes incredibly easy.
That’s why creating the English version above was even faster.
For AI,
language barriers barely exist anymore.
My Weekly Livestream Experiment
Every Sunday at 9 PM Korea time (5 AM Pacific Time)
I run a YouTube livestream.
I’ve been doing this for over a year now.
I explained the background of this livestream in a previous post:
50 Weeks of Vibe Coding — What I Built and What Comes Next
Each session usually lasts about three hours.
During that time I:
experiment with Vibe Coding using AI
explore new ideas and tools
learn in public
People who join the livestream
can also participate in the experiments.
The goal of the stream isn’t entertainment or even teaching.
It’s simply to experiment together.
The Problem With 3-Hour Videos
But there is one problem.
When the livestream ends,
you’re left with a 3-hour video.
To be honest,
even for me it’s difficult to rewatch the whole thing.
Occasionally I edited and uploaded short summary videos,
but that process required a lot of time and energy.
So I couldn’t keep doing it consistently.
As a result,
many interesting ideas and useful insights that appeared during those 3 hours
simply disappeared into the past.
I kept thinking:
What if AI could watch the entire livestream,
pick out the most important moments,
and automatically create a video summary?
The Moment That Idea Became Real
Last week I started a new series called:
“AI in Action: Experiments Applying AI in Everyday Life.”
During the first episode,
I was finally able to make that idea real.
This became possible thanks to:
AI4PKM / GOBI Desktop
AI4PKM is a project that explores how AI can be integrated with
PKM (Personal Knowledge Management).
And GOBI Desktop is an application designed to make the results of this project accessible to others.
What Happened During the Livestream
Before starting the livestream,
I launched the GOBI Desktop app and turned on the microphone feature.
While I was speaking during the stream,
AI4PKM’s Obsidian Canvas Skill organized my speech in real time.
It converted everything into a visual canvas structure.
Click the image below to jump directly to the relevant section of the YouTube video (the demo is in Korean).
After the 3-hour session,
everything I said had already been:
summarized
categorized
organized visually
as structured information.
What Happens Next
When the livestream ends,
I simply tell the AI:
Create a video based on this canvas.
Then AI4PKM’s markdown-video skill begins working.
The workflow looks like this:
Generate a script and slide structure from the canvas
OpenAI reads the script and generates the voice audio
Gemini creates slide images from the content
The audio and slides are combined into an MP4 video
The result:
3-hour livestream → summary video generated in about 10 minutes.
Once the Korean video is created,
generating an English version is almost trivial for AI.
If I want,
I could easily produce videos in many other languages as well.
The Beginning of the “AI in Action” Experiment
Most of the content in these livestreams is created using
my own learning system called VibeLearn AI.
When this experiment succeeded during the first episode of AI in Action,
I was genuinely excited.
Something that had previously existed only as an idea
had suddenly become real.
This Experiment Will Continue
Going forward,
I plan to continue using this workflow in my weekly livestream series:
“AI in Action: Experiments Applying AI in Everyday Life.”
Every Sunday
9 PM Korea Time
Each livestream itself will become another AI experiment.
As the system evolves,
AI should become even better at identifying the most valuable parts of long conversations
and transforming them into useful content.
I also plan to continue improving this workflow
together with the AI4PKM / GOBI Desktop development team.
Where This Technology Could Be Used
Right now I’m using this system mainly to:
summarize YouTube livestreams and instantly generate video content.
But I believe others could apply this technology
in many different ways.
The Core Tools I Use
The key tools behind my “AI in Action” experiments are:
VibeLearn AI
AI4PKM
GOBI Desktop
If you want to learn more about these experiments,
you can visit my website:
If you’re interested in participating in the AI4PKM project,
check out:
https://www.aiforbetter.me/index_en
You can also explore the AI4PKM web community, GobiSpace, here:
I will continue sharing these experiments.
And I hope we can explore new ways of applying AI in everyday life together.
It doesn’t matter what language you speak.
Everyone is welcome.
If this kind of experimentation interests you,
feel free to join anytime. 🙂
Thank you.