One Talk-head 을virtual 멀티 카메라 편집 으로 가져 오기Google Omni.

단일 카메라에 단일 클립을 촬영합니다. 이 기술은 당신의 연설의 비트를 찾아서 4개의 커트 타임스탬프를 가진 전투 시험한 신속한 템플렛을 채우고, Google Omni는 즉시 단단한 커트를 가진 새로운 사진기 각에서 당신의 진짜 영상을 재 짜맞춥니다. 프롬프트는 같은 얼굴, 방, 목소리 및 입술 동기화를 유지하기 위해 Omni를 요청합니다 (아래의 Known Omni limits 참조). 두 번째 카메라, 편집 타임 라인 없음.

어떻게 작동합니까?

1. 폴더에 이야기 헤드 비디오를 드롭하고 멀티캠 프롬프트를 요청합니다.
2. 기술은 ** 선명한 체크 ** (scripts/check_env.py])을 실행하고, 아무 것도 누락되는 경우, 그것을 설치하기 위해 권한이 무엇인지 설명합니다. 그래서 플러그 앤 플레이입니다.
3. 그것은 당신의 영상을 transcribes ** (Whisper, 로컬로 실행) 각 구문이 시작과 어디 숨을지 정확히 찾을 수 있습니다.
4. 그것은 **4 잘라 포인트 ** 그 비트에 (그들은 구문 시작, 내부 호흡, 결코 중학생) 그리고 왜 당신을 보여줍니다.
5. 당신은 타임 스탬프와 카메라 각도 (왼쪽 프로파일, 극단적 인 높은 각도, 닫기 업, 그리고 원래 튀기로 다시 - 그들 중 하나를 스왑하기 위해 미리 설정).
6. 명세.

Multicam

Turn ONE talking-head take into a virtual multi-camera edit prompt for Google Omni.

You film a single clip on a single camera. The skill finds the beats of your speech, fills a battle-tested prompt template with four cut timestamps, and Google Omni re-frames your real footage from new camera angles with instant hard cuts. The prompt asks Omni to keep the same face, room, voice and lip sync (see Known Omni limits below). No second camera, no editing timeline.

How it works

1. Drop your talking-head video in a folder and ask for the multicam prompt.
2. The skill runs a preflight check (scripts/check_env.py) and, if anything is missing, explains what it is for and asks permission to install it — so it’s plug-and-play.
3. It transcribes your video word-level (Whisper, runs locally) to find exactly where each phrase starts and where you breathe.
4. It suggests 4 cut points on those beats (cuts land on phrase starts, inside breaths, never mid-word) and shows you why.
5. You confirm or adjust the timestamps and the camera angles (left profile, extreme high angle, close-up, and back to the original framing — with presets to swap any of them).
6. It fills the frozen template, self-checks it (scripts/verify_prompt.py) and saves a ready-to-paste prompt. Upload your clip into Google Omni, select it as the source footage, paste, generate.

Setup (free and local, no API keys)

The skill checks this for you: scripts/check_env.py reports what’s missing, what each piece is for, and the install command for your OS — then asks before installing anything. To set it up manually:

• Install Python 3.8+.
• Install ffmpeg (includes ffprobe):
  • macOS: brew install ffmpeg
  • Windows: winget install Gyan.FFmpeg (or choco install ffmpeg)
  • Linux: sudo apt install ffmpeg
• Install the Python dependency: pip install -r requirements.txt

The first transcription downloads a Whisper model (~460MB for the default small) once. Everything else is local: no API key, no account, no paid service. Whisper is multilingual, so it works for any creator’s language (the delivered prompt is always English).

Install the skill

This is a Claude Code skill. Clone it into your skills folder:

git clone https://github.com/aipauloshimas/multicam ~/.claude/skills/multicam

(Windows: clone into C:\Users\\.claude\skills\multicam.)

Then open Claude Code, drop in your clip and say: “make the multicam prompt for this video”.

The template

The prompt template is frozen on purpose — its strict preservation rules (“do not alter the face… no changes to audio… only the camera position changes”) are what ask Omni to keep your identity, room and voice locked while the virtual camera cuts. The skill only ever fills the four [Xs] timestamps and, if you ask, swaps the angle descriptions. Everything else ships exactly as validated in production.

Known Omni limits