# KETAMI Studio Setup: turn Claude into a video editing machine

> **For the person:** open Claude Desktop, click the **Code** tab, choose **Local** and open an empty folder (for example `Documents/studio`). Drag this file into the chat, or type `@ketami-studio-setup.md`, and send: **"Follow this file."**
> This file only works where Claude can run commands: the Code tab of Claude Desktop, or Claude Code. In a normal chat, Claude can only read it.
> Needs a Claude Pro plan or higher.

---

## Instructions for Claude

You are setting up a video editing and motion design studio in the current folder. Follow the steps in order.

**Rules for this setup**
- Talk to the person in the language they use (Algerian Darija, French or English). Keep each message short.
- Before each step, say in one line what you are about to do. Ask before anything that needs admin rights, a password, or a large download (more than 500 MB).
- Never ask for or store API keys or passwords in files.
- Only install from the sources named in this file.
- If a step fails, explain the error in plain words, try the fix once, then move on and list it in the final report.
- Detect the operating system first and use the matching commands.

### Step 1: Check the basics
1. Detect the OS (macOS, Windows, Linux) and say which one you found.
2. Check that Node.js is installed (`node -v`). If it is missing, stop and ask the person to install the **LTS** version from https://nodejs.org, then continue.
3. Check for `ffmpeg`. If missing: macOS `brew install ffmpeg` (if Homebrew is missing, ask first), Windows `winget install Gyan.FFmpeg`, Linux the system package manager.

### Step 2: Remotion plugin (motion design)
- If the `claude` command works in the terminal, run:
  ```
  claude plugin marketplace add remotion-dev/claude-code-plugin
  claude plugin install remotion@remotion
  ```
- If it does not, tell the person to install it from Claude Desktop: **Code › Customize › Plugins › Discover**, search **Remotion**, install, then start a new session. Continue with the next steps meanwhile.

### Step 3: Remotion project
Create a Remotion project in this folder for vertical videos (1080x1920, 30 fps), using Remotion's official starter (blank template). Install its dependencies. Confirm that it starts.

### Step 4: auto-editor (silence cutting)
- macOS: `brew install auto-editor`
- Windows: download `auto-editor-windows-x86_64.exe` from the latest release at https://github.com/WyattBlue/auto-editor/releases, rename it `auto-editor.exe`, and put it in a `tools` folder here. Windows may warn "Unknown developer": the file is unsigned, that is expected.
- auto-editor is **no longer on pip**. Do not use pip for it.
- Test: `auto-editor --version`.

### Step 5: Local captions (Whisper)
1. Install `@remotion/install-whisper-cpp` and `@remotion/captions` in the project.
2. Install whisper.cpp with `installWhisperCpp` (pick one version, for example `1.5.5`, and pass the same `whisperCppVersion` to `transcribe` later).
3. On macOS, if the system asks to install developer tools, the person should accept.
4. Download the **multilingual `medium`** model (about 1.5 GB, ask before downloading). **Never** a model ending in `.en`: those are English only.
5. Whisper returns Arabic as word pieces. Always merge tokens into whole words with `createTikTokStyleCaptions` from `@remotion/captions` before showing captions.

### Step 6: Taste skills (Emil Kowalski)
Run:
```
npx skills add emilkowalski/skills --skill animate --skill animation-vocabulary --skill review-animations -a claude-code -g -y
```
This installs exactly 3 skills globally. Do not install the other 13.

### Step 7: Create CLAUDE.md
Create `CLAUDE.md` in this folder with exactly this content (the person will add their own fixes later):

```
# Editing rules
- Cut silences and dead air. Measure the silence level from each recording, never a fixed value.
- If I repeat a sentence, keep only the best take.
- Add a 30 ms audio fade at every cut so there are no clicks.
- Captions: word by word, centered, max 3 words on screen.
- Whisper splits Arabic words into pieces: merge tokens into whole words (createTikTokStyleCaptions) before showing captions.
- Arabic text: use a font that supports Arabic (Cairo or Noto Sans Arabic).
  Never animate Arabic letter by letter, it breaks the letters. Animate the whole word or line.
- Before the full render, render still frames and check them.
- One render at a time.
- End the video exactly on my last word. No black screen at the end.

# Anti-PowerPoint rules
- Every scene has depth: foreground, subject, background.
- Light is never flat: add a soft gradient, glow or shadow.
- The camera always moves a little (slow push or drift).
- Text is never centered by default. Use a real layout grid.
- Change the visual every 3 to 4 seconds.
```

### Step 8: Optional pro tools (ask first, both are optional)
- **earshot** (Claude can listen to audio). Needs Python 3.10+ and downloads about 1.1 GB of models. Install from https://github.com/Leonxlnx/earshot following its README, register it as an MCP server, and copy its two skills.
- **Openverse** (Creative Commons sounds). Add to this project's `.mcp.json` an HTTP MCP server named `openverse` with url `https://gateway.pipeworx.io/openverse/mcp`. It is a third-party server: remind the person to check each sound's license.

### Step 9: Final test and report
1. Open Remotion Studio and tell the person where to look (a browser tab).
2. Render one still frame of a 3-word Arabic title using a whole-word reveal, and show it.
3. Give a short report as a checklist: what is installed, what was skipped, what failed and how to fix it.
4. Tell them the next step: "Put a phone video in this folder as raw.mp4 and say: Edit raw.mp4 following CLAUDE.md."

---
Guide by Riyad Ketami (KETAMI). Checked against the official sources on 1 November 2026.