AI Makes My Videos — Once I've Said It, It's Done

Welcome back to Pinpin’s AI channel. Today’s episode was actually recorded yesterday. A lot of people watching our WeChat Channels have been asking:

You run this many projects at once, you’re extremely busy, you travel constantly — how do you ship a ten-to-twenty-minute long video every single day?

It’s actually pretty simple. Today I’ll show you how I do it with AI behind me.

1. Recording: OBS, don’t reinvent the wheel

Lots of creators have built their own screen recorders. What I want to say is — if you can find a good tool that already works, there is zero need to build your own. I use OBS.

I have one page configured (a bit of nesting involved):

• Webcam pulled into the frame
• Screen capture taking the left half
• Text stickers

Open the page, click start recording → hit stop when you’re done. Drop the resulting file into CapCut and speed it up, and today’s recording is finished. That simple.

2. Subtitles / covers / multi-platform copy — let AI do it

Then people ask: what about subtitles? Did you do them in CapCut? What about covers, and the synced copy for every platform?

Today I’ll share how AI does all of it.

I had Claude Code generate an HTML doc — just grab that doc from my site, hand it to your own Claude Code, and it can learn this whole methodology.

Four steps:

Step 1 — Transcript extraction

Pull the video transcript together (like I covered last episode), do the segmenting and translation, and turn it into a dozen-odd characters per line so it’s easy to read. Since I also publish on YouTube, it gets translated into an English version as well.

Step 2 — Subtitle burn-in

Not sidecar subtitles — a translucent band behind them that looks genuinely good, burned into the video.

Step 3 — Cover generation

Generated per my own methodology: real-person brand image + main headline + subheadline.

Step 4 — Per-platform copy

Following each platform’s requirements, differences, and moderation rules, write the copy for every platform’s post.

3. Every trap I’ve hit is in the doc

I moved quickly past how the SRT actually gets made and how recognition works — hand the doc to your agent and it can learn the process. I hit plenty of traps along the way:

• How to chunk the audio and run it concurrently for speed
• How to fit the platform ratios 3:4 / 9:16 / 16:9 / 4:3
• How to write copy that distills the content while carrying a mild clickbait edge to pull people in
• How to prompt image models like Nano Banana so the face stays consistent every time
• Each platform’s rules — WeChat Channels short titles can’t exceed 16 characters, and every other platform has its own length limit

Once you have the doc, your agent won’t fall into any of these traps.

4. What actually happens: one sentence

I already have CC (Claude Code) open, and I drop in the video I just recorded and tell it:

This is the video file I recorded today, already processed in CapCut. Please burn in the subtitles, make this episode’s cover, and generate a Markdown file with the distribution copy for every platform. When you’re done, run a check, and once it’s clean, tell me to review it and publish.

That’s really it. Every day, once I’ve said that sentence, I can go do something else.

It will:

• Put the finished cover in a folder on my computer
• Open the web file too
• Even open each platform’s publishing page for me

5. Why not fully automate publishing

Automating the publishing itself isn’t hard. But every platform has anti-scraping:

• Headless browsers are blocked
• WeChat Channels demands a QR-code login every single time

Tested in practice — rather than build an automation flow, it’s actually faster to have the page sitting open and click upload myself.

6. The final output

Task done, and it ran a self-check. A new folder for the episode in the working directory, split into: the knowledge video, the covers, the publishing page.

Covers: four versions, 3:4 / 4:3 / 16:9 / 9:16.

The publish page: titles, descriptions, and tags per platform, plus for WeChat Channels my own first comment on my own post, including the Chinese and English YouTube versions — all laid out. Extremely handy.

All I have to do is open all those platforms and post.

That’s my process for making a video every day — once I’ve said it, it’s done.

Hope there’s something in there for you.

Source: EP0014_audio.mp3 · ASR model gemini-3-flash (chunked parallel) · full text of the 5:30 original recording

[00:00] Welcome back to Pinpin’s AI channel. So this episode was actually recorded yesterday. The reason is that lately a lot of people watching our WeChat Channels account have been asking me: I run this many projects at once, work is extremely busy, and I travel constantly — so how do I manage to post a new dozen-odd-minute, even 20-minute long video every single day. And honestly this is pretty simple, so today I’ll show you how I get it done with AI backing me up.

[00:20] So first, I know a lot of creators have built their own screen-recording tools. Right, and what I want to say is that if we can find a tool that’s genuinely handy, genuinely good, ready to use — then there’s no need at all to build your own. And the one I use is a platform called OBS.

[00:34] So let me show you my page here, and it gets a bit recursive. Once I open this page, all I have to do is pull my camera into the frame, record my screen, capture the left half of my screen too, plus add a text sticker, and that’s the end of it. All I have to do is click start recording, and then hit stop recording when I’m done.

[00:52] The clip that comes out goes straight into CapCut, I speed up the parts that should be sped up, and today’s recording is done. Right, done — it’s that simple.

[00:59] And then people say, what about the subtitles? Right, are those done in CapCut? And also the cover image, plus the copy I post in sync across multiple platforms — how does that get written? So today let me share how my AI does all of this.

[01:15] So I had it generate an HTML web document as well. And don’t worry — as long as you get this web document off the official site and hand it to your own Claude Code, it can learn this whole methodology. And we can see it breaks into four steps.

[01:27] Step one is called full-transcript extraction. It takes my video’s full transcript — just like we covered last episode, it pulls the transcript together and does sentence segmentation and translation. It turns the subtitles into lines of a dozen-odd characters each, so they’re easy for viewers to read. And since I post to YouTube too, it also translates an English version.

[01:45] Step two is burning those subtitles into the video. They’re not external subtitles anymore — they become subtitles with a semi-transparent background, a really nice-looking effect.

[01:54] And step three is generating the cover following my methodology. The cover includes my real-person brand likeness, plus an attention-grabbing headline and subhead. And the last one is writing the copy for each platform’s post for me, based on each platform’s requirements and the differences between them, including some of each platform’s moderation principles.

[02:13] And the specifics of how it does the later stuff — how it produces the SRT file, how it does the recognition — we’ll move through quickly. Because once you hand this document to your agent, it can learn the process itself. And along the way I hit quite a few pitfalls, including how to segment things and run them in parallel to speed it up. And with this document in hand, your agent won’t hit any of those pitfalls at all.

[02:34] Including the aspect ratios for each platform, right — there’s 3:4, there’s 9:16, there’s 16:9, and there’s landscape ratios like 4:3, and how to adapt for all of that. Plus how, in the copy, to distill this episode’s content while still being just a little bit clickbaity to pull people in.

[02:50] Including how, when generating a face, you give it to these image models like Nano Banana in a way that gets it to keep the face consistent. Including what the standards are for post copy on each platform — for instance, WeChat Channels short titles can’t run over 16 characters, right, and every other platform has its own character limit, plus certain quirks of its own. And all of that gets factored in.

[03:09] So let me actually demo it for you. I’ve already opened CC. All I have to do is throw in the video I just recorded, and tell it:

[03:18] This is the video file I recorded today, already processed in CapCut. So I want you to burn subtitles into this video, and at the same time make this episode’s cover images for me, and generate a Markdown file with all the copy I need for distributing across every platform. And once all of that’s done, run one more check, and when the check comes back clean, ping me to review it and publish.

[03:45] Good, and it really is that simple. Every day, once I’ve said that line, I can go do something else. It puts the finished covers into a folder on my computer, opens all the web files too, and can even open the publishing page for each platform for me.

[04:00] This part isn’t hard to turn into automatic publishing. But every platform has some anti-scraping measures, including blocking headless browsers, or — like the WeChat Channels platform — requiring a QR-code login every single time. So from my actual testing, rather than building an automated flow, it’s actually faster to just have the page sitting open there and click upload myself.

[04:21] So let’s speed this process up, and once it’s finished in a bit I’ll show you the result. And I can go do other things.

[04:28] Good, it’s actually finished, it’s running a self-check. So let’s take a look. It’s already — I have a working directory, and under the working directory it’s created this episode’s folder for me, and split what’s inside into: task complete, the subtitled video, the covers, and this publishing web page.

[04:48] Alright, covers first. It made me a 3:4 version, then a 4:3 version, then this 16:9 version, and a 9:16 version too. We won’t open them one by one. And it also made that publish page from before, with the title for each different platform, its description, its tags.

[05:09] Plus, for WeChat Channels, my own first comment on my own post. Including the Chinese and English versions for YouTube, all of it pulled together. It’s extremely handy, extremely handy. All I have to do is open all those platforms and post it.

[05:20] So that’s my daily video-making process — basically once I’ve said it, it’s done. I don’t know whether that gives you any ideas. See you next episode, bye-bye.