mcps
E R D

Recipe

Turn a script into a captioned explainer

Narrate in ElevenLabs, generate matching scenes in Runway, then cut and caption it in Descript by editing text — a social-ready explainer with no camera.

for Anyone who wants a short video from a photo or a sentence — no editing software

Hand-checked recipe · curated by Ilya Gindin

~50 minutes

to run

intermediate

difficulty

3

connectors

What you'll need

ElevenLabs Reads your script in a natural presenter voice — the spine the whole video hangs on.
Required
Runway Generates the scenes and b-roll that sit under each line of the script.
Required
Descript Assembles voice and scenes, removes filler, and burns accurate captions — by editing the transcript, not a timeline.
Required

This recipe runs in 4 steps

  1. 1

    Narrate the script in ElevenLabs

    Say to your AI

    Narrate this in a warm presenter voice, conversational pace, natural pauses between sentences: '[paste your script]'.

    You'll get: One audio file with the full narration.

  2. 2

    Generate scenes sized to the lines

    Say to your AI

    Generate a 4-second cinematic scene for each of these beats, 9:16, consistent soft lighting and palette: [beat 1], [beat 2], [beat 3].

    You'll get: Short Runway clips, one per beat, in a consistent look.

  3. 3

    Assemble in Descript

    Say to your AI

    Import the narration and the clips, lay the scenes over the voice in order, and tighten any gaps.

    You'll get: A rough cut where each scene sits under its line.

  4. 4

    Caption and export

    Say to your AI

    Remove filler words and long silences, add centered captions from the transcript, and export a 1080x1920 MP4.

    You'll get: A captioned vertical explainer, ready to post.

You're done when

A captioned, social-ready explainer built from a written script — natural voice, matching scenes, accurate captions — with no camera, mic, or timeline editing.

The chain is ElevenLabs → Runway → Descript, about 50 minutes: ElevenLabs reads your script, Runway paints a scene per line, and Descript cuts the two together by editing text — filler words deleted like typos, captions generated from the same transcript so they’re accurate by construction.

That last seam is the one nobody documents: because the narration enters Descript as a transcript, the edit and the captions share one source of truth. No single tool’s docs cover it — ElevenLabs doesn’t know about Descript, Runway doesn’t know about either. This recipe is the hand-off.

Checked 2026-07-12: all three connectors reachable. ElevenLabs and Descript are official; Runway is official but installs as a local build.

If something breaks

I don't see Runway to turn on
Fix: Runway's connector is official but a local build: clone runwayml/runway-api-mcp-server, run npm install and npm run build, then add it with your RUNWAYML_API_SECRET.
I don't see Descript to turn on
Fix: In Claude: Customize → Connectors → search Descript → Connect (or add the custom URL https://api.descript.com/v2/mcp). In ChatGPT: add the Descript app.
Captions don't match the voice
Fix: They will if the voice went through Descript's transcript: captions are generated from the same text the edit runs on. Don't re-transcribe elsewhere — that's where mismatches come from.
Scenes and narration drift out of sync
Fix: Narrate first, then size each Runway scene to the sentence it covers. Matching scene length to its line is the whole sync trick.

People ask their AI

“make a captioned explainer from my script”“talking-head style video without filming”“script to social video with captions”

Related

ElevenLabs Descript turn a script into a narrated video