Generate faceless YouTube Shorts with n8n: script, voice and captions

[mina8]

Give n8n a niche and let it assemble the script, narration, footage and captions into a finished Short.

We built this at Zvid for narrated videos that don’t need a presenter on camera. The template connects OpenRouter, ElevenLabs and Zvid, with a manual trigger for testing and a daily schedule when you’re ready.

How it works

Channel brief → Script → Footage + narration → Captioned MP4 → Optional YouTube upload

1. Write the script. OpenRouter uses your niche and audience to produce narration split into scenes, plus a title and description.
2. Find the visuals. Each scene gets a stock search query. The workflow selects footage from Zvid’s stock library and looks for optional background music.
3. Generate the voice. ElevenLabs returns narration and alignment data, which the workflow converts into word timings.
4. Build and render. Zvid combines the scenes, voiceover, karaoke captions, transitions and channel branding into a vertical MP4.
5. Watch or upload. View the result inside n8n, or enable the included YouTube upload branch for future runs.

The timing part is useful to reuse: captions come from the same response as the voiceover, so there’s no separate transcription request. The video builder also uses those timings to calculate scene cuts and accounts for transition overlap.

Make it your own

Open Channel Config to change the niche, audience, voice, target duration, caption style, colors and outro. The Build project JSON Code node is where you can change the video layout further.

YouTube upload is off by default. When enabled, its default visibility is private. Add an approval step if you want a human checkpoint before uploading; review generated claims and footage before publishing.

To try it, install @zvid/n8n-nodes-zvid from Settings → Community nodes before configuring Zvid credentials; a workspace owner/admin may need to install it. Connect OpenRouter and ElevenLabs, then follow the setup notes on the canvas.

The template is free; service usage applies. The default dryRun: false renders and spends Zvid credits. Set it to true for an editor draft and render estimate, but OpenRouter and ElevenLabs still run and may charge.

Import the n8n template

Watch the walkthrough

Read the full setup guide

What would you add first: a Google Sheets script queue or an approval step before generating the voiceover?

[kartike]

Approval step first, and the reason is money. Your dryRun note says OpenRouter and ElevenLabs still run, so a bad script burns voice credits before you ever see the video. Put the checkpoint after the script node, send yourself the script with a quick approve or reject reply, and only approved rows reach the voice step. The Sheets queue second, and make it do double duty as the status log: script, approved, rendered per row, so a rerun never renders the same scene twice. Karaoke captions from the alignment data is a genuinely good trick, no second transcription pass.

[mina8]

Good point, especially about where the approval happens. The current dryRun skips the Zvid render, but the script and voice calls still run, so reviewing the finished draft is already too late to save voice credits.

A checkpoint after the script is parsed, with approve/reject before ElevenLabs, makes sense. I like the Sheets queue doubling as the status log too. Saving the voice URL and render job ID alongside the status would help reruns resume without repeating paid steps.

And thanks for noticing the captions! Reusing the ElevenLabs alignment for word highlighting and scene timing avoids that extra transcription pass. Appreciate the practical feedback.

[kartike]

no worries at all, im glad it helped
Gotta have a keen eye here ahha!