A repeatable recipe for making a 2–3 minute product vision video using AI tools, a copywriting framework, and code.

Total time: 15–25 hours across a weekend. Total cost: ~$25–50 in AI credits. Output: a shareable, on-brand vision video that reads like it took a month.

🎬 Demo: Watch Urban Pulse — Build a Relationship With Your City on Vimeo (2:44, made in one weekend with this recipe) Full walkthrough of the process behind that video: example/urban-pulse/

Why this exists

Most product vision videos have two problems:

  1. They cost $10k+ from an agency or take a month of your own time.
  2. When you DIY them, the seams show — inconsistent typography, uneven pacing, footage that doesn't match the script's emotional arc.

This kit fixes both. It combines:

  • A script framework (Ray Edwards' PASTOR) that guarantees a coherent emotional arc
  • AI tools for the expensive parts (voiceover, persona footage, script drafting)
  • Remotion — a video framework where you compose scenes in React, not a timeline editor. Everything is code, everything is version-controlled, everything is precise.
  • A centralized timing spine that keeps voiceover and visuals in lock-step even when you re-edit

The result: you can produce, iterate on, and re-render a 2–3 minute video as fast as you'd iterate on a landing page.

The 8-stage pipeline

1. SCRIPT              →  2. VOICEOVER         →  3. PERSONA CLIPS
   PASTOR framework       ElevenLabs               HeyGen (still → talking head)
   AI-assisted             ~$5                     ~$20-30

4. PRODUCT SCREENCAPS  →  5. MUSIC             →  6. REMOTION SCAFFOLD
   Playwright / Rewind      Uppbeat / Epidemic     Fork this template repo
   Free                     Free / $10             Free

7. SCENE COMPOSITION   →  8. RENDER & PUBLISH
   Timing spine → scenes    npx remotion render
   Iterate in Studio         YouTube / Vimeo

Each stage has its own doc under docs/. Follow them in order the first time; you'll only need 3 or 4 of them once you know the pattern.

What's in this repo

Folder Purpose
docs/ The playbook — one doc per stage of the pipeline
template/ A working Remotion scaffold — fork this to start a new video
prompts/ Copy-paste prompt libraries for HeyGen, ElevenLabs, and AI-assisted script writing
example/urban-pulse/ A fully worked example — script + timing map + final video reference

Quick start (30 minutes to your first render)

# 1. Fork/clone this repo
git clone https://github.com/yuxuant2025/vision-video-in-a-weekend
cd vision-video-in-a-weekend/template

# 2. Install
npm install

# 3. Preview in Remotion Studio
npx remotion studio
# → opens http://localhost:3000 with 8 stub scenes matching PASTOR beats

# 4. Render a 25% preview (fast)
npx remotion render VisionVideo --scale=0.25 --output=out/preview.mp4

# 5. Render full quality
npx remotion render VisionVideo --output=out/final.mp4

The template ships with 8 empty scene stubs — one per PASTOR beat — and a timing.ts spine. Start by dropping your voiceover mp3 in public/script/, updating the frame timings in timing.ts, and filling in scenes one at a time.

The framework: PASTOR

The script is the hardest and most important part. This kit uses the PASTOR framework by Ray Edwards, a copywriting framework originally designed for sales letters and email copy.

P — Person / Problem / Pain A — Amplify S — Story / Struggle / Solution T — Transformation / Testimony O — Offer R — Request / Response

Reference sources:

This kit's contribution: an adaptation of PASTOR from sales copy to 2–3 minute product vision videos, with the six beats mapped onto Remotion scene structure. See docs/01-script.md for the full mapping and worked example.

Every scene in the template/src/scenes/ folder is named after its PASTOR beat — so the framework lives in the code itself, not just in the docs.

What makes this different from a Remotion tutorial

Three things this kit does that a straight Remotion boilerplate doesn't:

  1. Centralized timing spine. All scene start/end frames live in one file: src/lib/timing.ts. Change your voiceover, edit one file, everything re-times. Most Remotion tutorials scatter timing across scenes; that's a nightmare when you re-cut.

  2. HeyGen ↔ product screencap interleave. The template's scene structure alternates persona-driven emotional beats with product-driven credibility beats. It's a compositional rule, not just a step. See docs/05-remotion-composition.md.

  3. Every AI tool has a prompt library. prompts/ has verbatim, copy-paste prompts for HeyGen persona briefs, ElevenLabs voice selection, and AI-assisted script drafting. You're not starting from a blank page anywhere in the pipeline.

Made by

@yuxuant2025

Built during the making of Urban Pulse — a vision video for an AI companion app that gives public spaces a personality, memory, and voice.