Making videos with AI through Claude, as a total beginner

I'm not a video editor. I just wanted to see whether Claude could actually make a video for me.

The first clip wasn't good. I'll show it anyway.

The first clip

This is the very first one. Text pops in, no sound, stiff timing.

About fifteen minutes later I had this. A bit of motion, but still a slide that moves.

Then I added a free voice. The picture barely changed, but the clip suddenly felt alive. Sound matters way more than I expected.

Where it is now

A few weeks on, it looks like this. A short excerpt from a recent clip.

I didn't suddenly get good. I just fixed one thing at a time. See something you don't like, tell Claude that one thing, repeat.

Want to try? Paste this

Install Node.js first (nodejs.org). Then open Claude Code, or ask in claude.ai and copy the code over yourself. Paste:

Make me a vertical 1080x1920 video, about 15 seconds, using Remotion.

1. Set up the project with npx create-video@latest
2. Three scenes, about 5 seconds each, with large readable text.
   The text should ease or slide in, not just appear.
3. The topic is "[what you want to talk about]".
   Write a short 3-sentence script in spoken Thai.
4. Generate a voice file from the script with edge-tts, voice th-TH-NiwatNeural,
   and time each scene to match the audio.
5. Add the audio to the video and render an mp4.
   If anything breaks, fix it yourself and tell me what you changed.

Want a female voice? Swap in th-TH-PremwadeeNeural.

Don't expect the first render to look good. Mine didn't. Watch it, then ask for one fix at a time: "text is too small", "scene two is too fast".

On the voice: edge-tts is free because it pulls from Microsoft Edge's speech service. I'd treat it as fine for personal use and learning. If you're making commercial content, use a TTS service whose licence clearly allows that.

What I hit between the first clip and now

The prompt above gets you a first clip. Getting from there to something I was happy to post took me a long time, mostly on these:

• The AI voice mispronounced brand names, every single clip
• The voice came out choppy, with dead air that screamed "AI"
• Mixed Thai font styles that looked amateur
• Visuals that didn't land on the words, fixed one scene at a time
• Hooks that didn't stop anyone scrolling
• And all of it again, every day

Each one cost days before it looked like the latest clip above.

If you'd rather not fix it all yourself

I packed everything I actually use into Captog Cue, which runs with your own Claude.

• Long video to shorts: picks the best moments and adds subtitles
• Daily: turns an interesting news item or tool into a Thai short with a caption, every day. It comes with Cue at no extra cost

The latest clip above was made this way. See the details, or comment DAILIES under a clip on the pemlog page and I'll send you the details.

Thanks to Remotion and edge-tts for being open source.