Open Source · AGPL-3.0

AICW Video

AICW Video is an open-source AI agent for editing video interviews with humans. It auto-matches and syncs a separately recorded audio track, generates captions, suggests short clip moments, blurs faces or replaces them with emoji, and can replace the speaker’s voice with computer-generated speech. Live preview for every option. Works as a standalone app and as a skill/MCP for Claude, ChatGPT and Codex.

  • Privacy-first
  • Live preview
  • Standalone + MCP
  • Claude / ChatGPT / Codex
  • Local-first
  • macOS
  • AGPL-3.0
GitHub → brew install aicw-io/tap/aicw-video
AICW Video editing a conference talk: speaker's face is blurred for privacy, captions are styled, and the moments-and-captions panel lists clip suggestions

Features

Auto-matches and syncs separately recorded audio

Drop video files and one or more separately recorded audio tracks. AICW Video detects, matches, and syncs each track to the right clip automatically.

Auto-generates captions

Speech-to-text captions in multiple styles. Edit the wording, switch styles, and see the result live before rendering.

Privacy: blur faces or replace with emoji

Detects faces frame by frame and either blurs them or covers them with a chosen emoji. Useful for interviews where the subject wants to stay anonymous.

Privacy: replace voice with TTS

Replace the original speaker’s voice with a computer-generated voice while keeping the captions accurate. Pick a voice, regenerate, preview.

Suggests short clip moments

Analyzes the interview and proposes the strongest short ranges to cut. Accept, reject, or adjust each suggestion.

Live preview for every option

Caption styles, face emoji, voice replacement, clip ranges — every choice updates the preview instantly. No render-to-check round trips.

Standalone app + skill/MCP

Use the local hub on your own, or drive AICW Video from Claude, ChatGPT, or Codex via its built-in MCP server.

Caption silent video

No usable audio? AI scene analysis describes what’s on screen so you can ship a captioned clip anyway.

See it in action

Privacy demo: blur the speaker’s face, replace it with an emoji, and replace the original voice with TTS — all with live preview.

Clip workflow: auto-sync separately recorded audio, get suggested clip moments, generate captions, render the final cuts.

Screenshots

Privacy modes, plan UI, and the multi-project hub.

AICW Video editing a conference talk with the speaker's face blurred and yellow styled captions, with the moments-and-captions panel on the right
Privacy mode: blurred face + live caption preview.
AICW Video editing the same conference talk with the speaker's face replaced by a circular emoji placeholder while captions remain accurate
Privacy mode: face replaced with an emoji.
AICW Video plan UI showing the source video, a timeline of suggested clip moments, and the clip list
Plan UI: source video, suggested moments, clip list.
AICW Video local hub listing project plans with status badges and Open plan UI buttons
Local hub: every project on one page.

Drive it from Claude, ChatGPT or Codex

AICW Video ships a local stdio MCP server, so it works as a skill inside Claude Code, Claude Desktop, ChatGPT, and Codex. The host can import a video, analyze it, create a plan, and render clips on your behalf — the local app stays the source of truth for previews and final renders.

Add to Claude Code:

claude mcp add aicw-video -- aicw-video mcp

Then prompt:

use aicw-video to cut /path/to/video.mov into clips
Claude Code calling AICW Video to create and analyze a project

Install

From Homebrew (recommended). Pulls Node.js, ffmpeg-full, and whisper-cpp as dependencies:

brew install aicw-io/tap/aicw-video

Or the development build from the upstream main branch:

brew install --HEAD aicw-io/tap/aicw-video

Then start the local hub:

aicw-video

The browser hub opens at http://127.0.0.1:8764/. macOS is the supported platform today; Windows support is planned.

How AI is used

AICW Video processes locally: audio and video extraction, audio-to-text (Whisper), and face detection (TensorFlow).

For optional AI scene analysis it uses the AI tools you already have installed — Claude Code, Codex CLI, or a local Ollama model. When that's on, sampled frames and transcript snippets may be sent to the host you choose. When it's off, AICW Video stays fully local.

Need full local AI? Configure Ollama with a local model like Qwen or Gemma.

Loading PDF…