Tonight, I was watching a random video on YouTube – nothing too special – until I clicked on their website... and there was absolutely nothing related to the video content.

And just like that, an idea hit me:
Why not build an app that turns YouTube videos into articles?
Sounds kinda legit, right?

Where to begin?
I started with the YouTube Data API v3 to fetch transcripts.
But yeah... dream on.
Turns out it only lets you fetch transcripts from videos you actually own, and it takes like three different API calls. Kinda messy. So I switched gears and found a Python package called youtube-transcript-api. No authentication needed, and it worked like a charm.

Boom – Step 1: Get the transcript.

Next up: use AI to convert the transcript into a full-blown article.
But if you’ve ever looked at YouTube auto-generated transcripts, you know they’re often a hot mess. Like, in a video about “Phú Quốc,” the transcript came out as “fuk quak.” So before feeding it to AI, I had to clean the data:

• Fix spelling errors

• Check grammar

• Remove weird characters

• Merge or delete irrelevant segments

• And most importantly: add context (like places, people, and overall topic)

Only after all that can the AI turn it into something actually worth reading. But I wasn’t done yet. To make the article pop, I needed images from the video. At first, I thought of using a headless browser to play the video and take screenshots.
But then I hit roadblocks:

• Ads

• Laggy internet

• And the images weren’t even sharp or on-point

So now I’m leaning towards downloading the video and extracting still frames directly. Seems like a cleaner route. But hey, it’s getting late. I’ll deal with that part... tomorrow.