Behind the scenes of Mark học AI: how AI makes and posts the videos

Special episode #100

This is a special episode: video number 100 of the channel. Mark wonders: how is all of this made? Bit answers: let's open up the backstage. Every video is made by AI together with one person.

STEP 1

It starts with a prompt

Everything starts with a prompt, which is just a message. Claude writes the script and splits it into scenes, one idea per scene. But nothing happens next until the person running the channel approves the script.

STEP 2

A person picks the logo

Same with the logo. Claude drew 9 designs, and the person running the channel picked one: an M made from the letters A and I. Then it changed to sky blue, also someone's choice.

STEP 3

An AI voice

The voice you're hearing is an AI voice from ElevenLabs. There were 8 male voices to try, and this one was chosen. The whole script is read in one go, then cut into sentences, so the voice doesn't change halfway.

STEP 4

Drawing with code

The pictures aren't filmed. Claude writes code that makes the computer draw each frame, 30 frames per second, matched to each spoken sentence. Then it exports two versions: vertical for phones, horizontal for YouTube.

A WHOLE TEAM

A whole team of Claudes

And it's not just one Claude. One Claude coordinates, and each series gets its own Claude, all working in parallel. That's multi-agent. When they're done, another AI listens back to every sentence with speech-to-text to catch misread words.

FIX MISTAKES

The logo got covered

And it isn't always right the first time. Once, the phone's on-screen buttons covered the logo. The person running the channel sent back a screenshot, Claude fixed it, and since then every video stays inside the safe area.

STEP 5

Posting through Metricool

Finished videos go into Google Drive. Claude grabs the links, writes titles and descriptions, then schedules everything in Metricool: 5 short videos a day for YouTube, TikTok and Facebook, plus one long video that combines a whole series.