THE BLACKCAT Video Vision Studio

# THE BLACKCAT Video Vision Studio

THE BLACKCAT Video Vision Studio wraps `claude-video-vision@1.3.2` inside a local dashboard for video inspection, frame extraction, metadata and offline transcription.

## Local Dashboard

- URL: `http://127.0.0.1:8772`
- Launcher: `scripts/open-video-vision-studio.command`
- Studio folder: `studios/video-vision-studio/`
- Input folder: `studios/video-vision-studio/input/`
- Output folder: `studios/video-vision-studio/output/`
- Reports: `studios/video-vision-studio/output/_reports/`
- MCP package: `claude-video-vision@1.3.2`

## What It Does

1. Uploads or reads a video from the local `input/` folder.
2. Calls `claude-video-vision` through MCP using the local Node helper.
3. Configures the backend as local Whisper by default.
4. Extracts a sampled frame set as JPG.
5. Saves JSON and Markdown reports under `output/_reports/`.

Default settings:

- Frame sample: 18 frames.
- Frame width: 512 px.
- Backend: local Whisper.
- Whisper model: `tiny` for first-pass speed.
- Audio can be skipped from the dashboard.

## One-Time Setup

The launcher installs the local Node dependencies automatically on first open:

```bash
scripts/open-video-vision-studio.command
```

Manual setup:

```bash
cd studios/video-vision-studio
npm install --omit=dev
```

Required local tools:

```bash
brew install ffmpeg yt-dlp whisper-cpp
```

This Mac already had `ffmpeg`, `ffprobe`, `yt-dlp` and `whisper-cli` available when the Studio was added.

## How To Use

1. Open THE BLACKCAT Video Vision Studio from the portal.
2. Upload a video or press `Abrir input` and place a video there.
3. Press `Actualizar`.
4. Choose frame count, resolution and Whisper model.
5. Press `Analizar`.
6. Review the frames, transcript and saved report in `output/_reports/`.

## Privacy Notes

- The dashboard configures `claude-video-vision` with local Whisper by default.
- Do not use Gemini or OpenAI backends for sensitive client material unless the operator explicitly decides that the third-party API is acceptable.
- YouTube URLs can be supported by `claude-video-vision`, but this dashboard is intentionally local-file-first.
- Do not commit videos, frames, reports, transcripts, `node_modules`, logs or local config.

## MCP Bridge

The bridge script is:

```bash
studios/video-vision-studio/scripts/mcp-call.mjs
```

It launches the MCP server and calls:

- `video_configure`
- `video_watch`
- `video_setup`
- `video_info`

The package is pinned in:

```text
studios/video-vision-studio/package.json
```