Make a YouTube Shorts video with AI

Every "make a Short with AI" tutorial hands you the same four tools and calls it a day. What none of them tell you is where the workflow actually breaks — the voice that sounds robotic, the video clips that don't match the script, the captions that get cut off on mobile. This walkthrough is the version after you've made a few and learned where the friction is. Follow it and you'll have a finished, captioned, upload-ready Short in about 30 minutes, for $0–$10.

The pipeline is deliberately linear: script first, because everything downstream needs it; voice before video, so you can time the visuals to the narration; captions last, because they're the cheap step that most affects watch time. Skip the order and you'll redo work.

What you'll walk away with

• A 45–60 second script written for retention, not just information
• A natural-sounding AI voiceover
• AI-generated vertical video clips that match the script
• Burned-in captions in a style that survives muted autoplay

Step by step, with the judgment that matters

1. Script with ChatGPT — the first three seconds are everything. Ask for a 45–60 second script that hooks in the first three seconds, uses short sentences, and ends with a call to action. Most Shorts are lost in the opening line; make the model rewrite the hook until it earns the next five seconds. Ask for spoken lines only — no stage directions muddying the voiceover step.

2. Voice with ElevenLabs — drop the stability slider. Paste the script, pick an energetic voice, and set stability around 40%. Higher stability sounds flat and robotic; lower gives the natural emphasis that keeps people listening. This one slider is the difference between "AI voice" and "voice."

3. Video with Kling — describe scenes, not the whole video. Generate vertical (9:16) clips scene by scene from your script, asking for cinematic, high-detail, smooth motion, no text overlays. Generating per scene gives you clips you can actually cut to the narration; one long generation rarely lines up.

4. Captions with CapCut — bold, high-contrast, and safe from the edges. Import clips and voiceover, run Auto Captions, and pick a bold high-contrast style. Most Shorts are watched muted, so captions aren't optional — they're the content. Keep them clear of the bottom UI zone and export at 1080×1920.

A real run

For a Short on "why your houseplant is dying," ChatGPT opens with "You're probably killing it with kindness" — a hook that survives the swipe test. ElevenLabs voices it warm and quick at 40% stability. Kling generates three clips (a drooping plant, a watering can, a happy plant) that you cut to the narration beats. CapCut auto-captions in bold white, and you nudge them up off the bottom bar. About 30 minutes, $0 because Kling's free credits covered three short clips.

Where it goes wrong

• The voice sounds robotic. Stability's too high or the voice is wrong. Drop to ~40% and try another voice.
• Clips don't match the script. You generated one long video. Generate per scene and cut to the voiceover instead.
• Captions get cut off. They're too low. Raise them above the bottom UI zone before exporting.

The honest cost

$0 to about $10. ChatGPT, ElevenLabs, and CapCut free tiers cover one Short comfortably. Kling's free credits are the only limiter — budget a few dollars if you regenerate video clips a lot. Swap Kling for Runway, or ChatGPT for Gemini, and the workflow is identical.