explaingit

kiendt91/video-mix-with-codex

22PowerShellAudience · vibe coderComplexity · 3/5ActiveSetup · hard

TLDR

Windows PowerShell workflow that combines Codex, FFmpeg, and HyperFrames to plan and render beat-cut or cinematic music videos from a JSON edit decision list.

Mindmap

mindmap
  root((video-mix-with-codex))
    Inputs
      Music file
      Source clips
      Edit decision JSON
    Outputs
      Rendered MP4
      Contact sheets
      Quality reports
    Use Cases
      Beat-cut music video
      Cinematic story video
      Re-render with cache
    Tech Stack
      PowerShell
      FFmpeg
      HyperFrames
      Node.js 22
      Studio UI

Things people build with this

USE CASE 1

Edit a beat-cut music video where cuts land on song beats

USE CASE 2

Build a cinematic story music video that weaves dialogue and shots over a track

USE CASE 3

Render an MP4 from a JSON edit decision list with automated black-frame and subtitle checks

USE CASE 4

Use the local Studio UI to plan edits with optional AI providers like OpenAI or Anthropic

Tech stack

PowerShellFFmpegHyperFramesNode.jsCodex

Getting it running

Difficulty · hard Time to first run · 1h+

Windows-only, requires FFmpeg, FFprobe, Node 22, HyperFrames CLI on PATH, and optionally AI provider API keys stored in APPDATA.

In plain English

This project is a workflow for editing music videos on a Windows machine, combining several pieces of software. The author uses Codex (an AI coding assistant), FFmpeg (a command-line tool for cutting and converting video), and HyperFrames (another command-line tool that turns an edit description into a rendered video). The repository is mostly a collection of PowerShell scripts and example files that glue these pieces together. The README explains the philosophy first. Instead of picking video timestamps at random, the author wants the editor to listen to the song, look at every clip of footage, decide what each shot is for, then render and check the result before delivering it. It supports two edit styles. A beat-cut music video where cuts land on the beats of the music, and a cinematic story music video where dialogue and character shots tell a small story over the song. The core process is described in six stages: prepare the media files, inspect each source clip for things like episode cards or hard subtitles, build contact sheets that show frames from each clip, plan the story or the song timing, mark which audio belongs to which moment, then generate, render, and run a quality check. An edit decision list, written in JSON, drives all of this. The scripts validate the list, build a HyperFrames project from it, render the final MP4, and run automated checks for black frames or accidental subtitles. To run the workflow you need Windows PowerShell, FFmpeg and FFprobe on your PATH, Node.js 22 or newer, and the HyperFrames command-line tool. There is also a local web interface called the Studio UI. Running npm run studio opens a browser page at a local address where you can paste paths to your music, source videos, and an output folder, then trigger each step in the pipeline from buttons rather than the command line. The Studio UI optionally talks to AI providers like OpenAI, Anthropic, Gemini, or xAI to draft, repair, or review the edit decision list. API keys are stored outside the repository in an APPDATA folder, and the actual rendering stays local with FFmpeg and HyperFrames. The README provides two quick-start sequences, one for each edit style, and a faster pipeline script for repeat jobs that caches normalized clips so re-renders skip unchanged shots.

Copy-paste prompts

Prompt 1
Walk me through installing FFmpeg, HyperFrames, and Node 22 on Windows so I can run this video-mix workflow
Prompt 2
Generate an example edit decision JSON for a beat-cut music video using three source clips and a 3-minute song
Prompt 3
Start the Studio UI on my machine and show me how to paste paths for music, source video, and output folder
Prompt 4
Run the quality-check stage on my latest render and report any black frames or hard subtitles found
Prompt 5
Set up the fast pipeline script so repeat renders reuse cached normalized clips
Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.