ScriptCutAI

AI-powered video editing assistant that automatically removes mistakes, retakes, and off-script talking from script-based recordings.

What It Does

• Transcribes your video using AI (Whisper)
• Compares what was said to your script
• Finds the best take for each sentence
• Removes mistakes, pauses, and off-script talking
• Applies cuts directly in Premiere Pro

How It Works

Premiere Pro panel (CEP: HTML/JS + ExtendScript)
      │  1. picks the clip on the active timeline, sends its path + your script
      ▼
Local FastAPI server (Python, http://127.0.0.1:8000)
      │  2. Whisper transcribes the audio with word-level timestamps
      │  3. Alignment finds every attempt at every script sentence
      │  4. Cut engine picks the best take per sentence, in script order,
      │     and trims long silences inside kept takes
      ▼
Edit list (KEEP / REMOVE segments with timings)
      │  5. ExtendScript razor-cuts the sequence at segment boundaries and
      │     ripple-deletes the REMOVE parts (or just adds markers, as a fallback)
      ▼
Clean, script-ordered edit on the timeline

• Transcription (backend/app/services/transcription.py): OpenAI Whisper runs locally on the CPU (default model base; tiny through large are selectable). It returns word-level timestamps. The model stays loaded between runs and can be unloaded on demand.
• Alignment (backend/app/services/alignment.py):
  • The script is split into sentences, and the text is normalized.
  • The transcript is scanned for attempts at each sentence using Levenshtein similarity on key words.
  • Filler words and off-script talking are recognized, so retakes, false starts and chatter can be told apart.
• Cut decisions (backend/app/services/cut_detector.py):
  • Selects the best take for every sentence while enforcing the script's order.
  • Generates KEEP/REMOVE segments with tight boundaries, merges neighbours and removes long internal silences.
  • Reports how much time is kept and how much removed.
• Premiere integration (premiere-extension/jsx/premiere.jsx):
  • Reads the active clip.
  • Converts seconds to Premiere ticks.
  • Applies cuts with razor + extract through the QE DOM, with a marker-only fallback plus marker clearing and seeking helpers.

Tech Stack

Part	Technology
Backend	Python 3.12, FastAPI + Uvicorn, Pydantic
Speech-to-text	OpenAI Whisper (local, PyTorch CPU build), FFmpeg
Text matching	python-Levenshtein, RapidFuzz
Editor panel	Adobe CEP extension (HTML/CSS/JS, CSInterface) + ExtendScript for Premiere Pro (host PPRO 13.0+)
Setup	One-click INSTALL.bat → setup-new-computer.ps1 (installs Python if needed, creates the venv, installs dependencies, installs and enables the extension)

API

Method	Endpoint	Purpose
GET	/health	Server status (the panel shows "Server offline" when this fails)
POST	/api/transcribe	Transcribe a video file with word-level timestamps
POST	/api/analyze	Full pipeline: transcribe, align to the script, return the edit segments and statistics
GET	/api/edits/premiere-format	Describes the edit-list format the panel consumes
POST	/api/unload-model	Free the Whisper model from memory

Interactive API docs are served by FastAPI at http://127.0.0.1:8000/docs while the server runs.

Requirements

• Windows 10/11
• Python 3.12
• Adobe Premiere Pro (CC 2019 or later)
• ~2GB disk space for AI models

Installation (One-Time Setup)

Quick option: double-click INSTALL.bat. It runs setup-new-computer.ps1, which:

• installs Python if needed,
• creates the backend's virtual environment and installs the dependencies (it stops with an error if that fails),
• installs and enables the Premiere Pro extension.

Or follow the manual steps below.

Step 1: Install Python 3.12

Download from: https://www.python.org/downloads/

• Check "Add Python to PATH" during installation

Step 2: Install FFmpeg

Open PowerShell and run:

winget install Gyan.FFmpeg

Step 3: Set Up the Backend

Open PowerShell in the ScriptCutAI folder and run:

cd backend
python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt

Step 4: Install the Premiere Extension

Run as Administrator:

.\install-extension.ps1

Or manually:

1. Copy the premiere-extension folder to: C:\Users\YOUR_USERNAME\AppData\Roaming\Adobe\CEP\extensions\com.scriptcutai.panel

2. Enable unsigned extensions (run in PowerShell):

reg add "HKCU\Software\Adobe\CSXS.11" /v PlayerDebugMode /t REG_SZ /d 1 /f

Usage

Step 1: Start the Server

Double-click start-server.bat or run:

cd backend
.\venv\Scripts\Activate.ps1
uvicorn app.main:app --host 127.0.0.1 --port 8000

Step 2: Open Premiere Pro

1. Open your project with the video
2. Go to Window → Extensions → ScriptCutAI

Step 3: Analyze and Cut

1. Click "Get from Timeline" to select your video
2. Paste your script text
3. Click "Analyze Video" (wait for transcription)
4. Click "Preview Cuts" to review
5. Click "Apply Cuts to Timeline" to execute

Folder Structure

ScriptCutAI/
├── backend/              # Python AI server
│   ├── app/              # Application code
│   ├── venv/             # Python environment (created by setup, not in the repo)
│   └── requirements.txt  # Dependencies (pinned; uses PyTorch's CPU wheel index)
├── premiere-extension/   # Premiere Pro panel
├── start-server.bat      # Quick start script
├── install-extension.ps1 # Extension installer
└── README.md             # This file

Troubleshooting

"Server offline" in the panel

• Make sure you started the server first
• Check if port 8000 is available

Extension not showing in Premiere

• Run the install script as Administrator
• Restart Premiere Pro completely

Analysis takes too long

• Use "tiny" or "base" Whisper model for faster results
• "medium" is more accurate but slower

Cuts not applying

• Make sure a sequence is active
• Try the "Add Markers" fallback option

Tips

• Keep the server running while editing
• Use "base" model for good balance of speed/accuracy
• Review cuts with "Preview" before applying
• Save your project before applying cuts!

Repository Notes

Changes made when preparing this repository (the working copy on my PC is unchanged):

• backend/requirements.txt was converted from UTF-16 to UTF-8 so GitHub and every tool can read it. It now starts with --extra-index-url https://download.pytorch.org/whl/cpu, because the pinned torch==2.9.1+cpu builds are only published on PyTorch's own index. Without that line, pip install -r requirements.txt fails on a fresh machine.
• setup-new-computer.ps1 no longer hides pip errors (2>$null). If installing dependencies fails, it now stops and says so, instead of reporting success.
• install-extension.ps1 prints the backend path relative to where the script lives, instead of a hardcoded personal folder.
• The Python virtual environment (backend/venv, about 1 GB) is not stored in git; setup recreates it.

Copyright © Omer Avcioglu (McHunter Studio). All rights reserved. Viewing only; see LICENSE.