Abhishek Sira Chandrashekar

OpencutAI shipped an AI Co-Pilot that edits videos for scene detection and YouTube chapters.

Hey everyone!

Three major features just landed in OpenCut AI and they all run locally on your machine.

1. AI Co-Pilot Agent

This is the big one. Tell the editor what you want in plain English and it creates a step-by-step plan, then executes it.

Examples:

- "Make this a 60-second vertical reel with captions and trending music"

- "Remove all silences and filler words, then normalize audio for YouTube"

- "Add cinematic color grading and auto-duck the background music under speech"

How it works:

1. You describe your goal

2. The AI analyzes your current project (tracks, clips, transcript, duration)

3. It returns a multi-step editing plan with 19 possible action types

4. You review each step, then click Execute

5. Steps run sequentially with live progress — cancel anytime

We added 6 quick presets for common workflows. The whole thing runs through your local LLM — no cloud, no data leaving your machine.


2. Client-Side Scene Detection

Detect visual scene changes (camera cuts, angle changes, lighting shifts) without sending a single frame to a server.

How it works:

- Extracts frames at configurable intervals (default: every 0.5s)

- Computes 24-bin RGB color histograms for each frame

- Calculates chi-squared distance between consecutive frames

- Flags changes above threshold as scene cuts or dissolves

- Shows before/after thumbnails for each detected scene change

All processing happens in the browser using Canvas API. No backend needed.


3. YouTube Chapters Export

We already had auto-chapter detection (AI analyzes your video and finds topic boundaries). Now you can export them directly in YouTube's `MM:SS Title` format:

```

0:00 Introduction

1:23 Why Privacy Matters

3:45 How to Self-Host

7:12 Q&A

```

One click copies to clipboard. Also generates a full YouTube description with your suggested title, description, chapter timestamps, and tags.


Why this matters

Every feature runs locally. Descript sends your video to AWS. CapCut sends it to ByteDance. Runway sends it to their cloud. OpenCut AI keeps everything on your hardware — your footage, your AI models, your rules.

Self-host on a $20/mo server or run it on your laptop. MIT licensed.



What would you want the Co-Pilot to do next? 🎬

272 views

Add a comment

Replies

Best
Callum Robert

@humblefool60 Does the local AI need a beefy GPU to run the Co-Pilot, or does it work okay on a standard MacBook?

Abhishek Sira Chandrashekar

@callum_robert It is better to have but works without it too (but slower)

Wesley Arthur

@humblefool60 Can the Co-Pilot actually handle complex edits like multi-track syncing, or is it mostly for quick social media clips?

Abhishek Sira Chandrashekar

@wesley_arthur Yes, it will be. I'll be adding more features into it.

@humblefool60 If I'm running this locally, does it tend to heat up the laptop quite a bit during the "Execute" phase?

Abhishek Sira Chandrashekar

@kinsley_brooke Sadly, a little bit of computing will be used that can trigger the heat that can trigger a heating in the laptop.

@humblefool60 How does the auto-ducking feel—does it sound natural or do the volume shifts feel a bit robotic?

Adrian Cole

@humblefool60 If I cancel a step mid-way through the execution, does it revert the project or just stop where it is?

Abhishek Sira Chandrashekar

@adrian_cole4 Sadly, no, there is no context for the thread that has been preserved, but it goes back to zero, so you have to start over once again.

Jaimy James

For me, the biggest win is the local AI processing. I work with sensitive footage sometimes, so keeping everything on my machine is a huge advantage over cloud editors.

James Darren

Honestly, the direction is strong especially the pain English structured edit plan → user approval → execution” loop.

If you extend the Co-Pilot, the next big win would be multi-video story assembly (auto picking best clips across folders/projects) or auto social variants (TikTok vs YouTube vs Shorts in one click) 🚀

Julie Grayson

Next step I’d want is multi clip intelligents like the Co-Pilot automatically finding the best moments across long footage and stitching them into a first rough cut. After that platform specific export (TikTok/Shorts/YouTube) in one click would be huge 🚀

Abhishek Sira Chandrashekar

@julie__grayson Exactly, this is the direction we are looking at.

Paisley Coleman

@humblefool60 Does the auto-zoom for the spotlight get confused if there are multiple people or faces in the frame?

Abhishek Sira Chandrashekar

@paisley_coleman This is an interesting question. I haven't tried it yet, but definitely the testing results would be interesting.

Rivra

Running locally is a huge plus for video editing because privacy and speed are key. How do you handle the processing power requirements for running those models locally on a standard laptop versus a high end workstation? Great work on the scene detection feature!

Abhishek Sira Chandrashekar

@rivra_dev I am running the model on my local system. If you have a better processor with a GPU, then it's going to be much simpler and faster.

12
Next
Last