# reel-setup

Adopt this skill: https://proskills.md/skills/other/jenesyx-arta-edit-studio-claude-skills-reel-setup

Page: https://proskills.md/skills/other/jenesyx-arta-edit-studio-claude-skills-reel-setup
This document is the public skill bundle for that page. It contains SKILL.md and every helper file the catalog carries. No login is required.

## File: .claude/skills/reel-setup/SKILL.md
size: 3583
sha: 0d318e911bd362882384ce1568d9aa1885bac867
upstream: https://github.com/jenesyx/arta-edit-studio/blob/main/.claude/skills/reel-setup/SKILL.md

```text
---
name: reel-setup
description: Use when the user runs /reel-setup, has just cloned Arta Edit Studio, or a reel command fails because Python packages, Node packages, ffmpeg, fonts or reel.config.json are missing.
argument-hint: (optional) your name, Instagram handle, brand
---

# Arta Edit Studio — setup

Get a fresh clone ready for `/reel-story` and `/reel-split`, then personalise it.
Explain each step in one plain sentence before running it — many users are not
developers. **Ask before anything that downloads or installs** (packages, fonts,
model weights) and say roughly how big it is.

Arguments: `$ARGUMENTS`

## 1. Check the machine — report as a ✅ / ❌ table

| Need | Check | Fix if missing |
| --- | --- | --- |
| Python ≥ 3.10 | `python --version` | python.org |
| Node ≥ 18 | `node --version` | nodejs.org (LTS) |
| ffmpeg **with zscale** | `ffmpeg -hide_banner -filters` contains `zscale` | Windows: gyan.dev "full" build · macOS: `brew install ffmpeg` |
| NVIDIA GPU (optional) | `nvidia-smi` | none → Whisper runs on CPU, ~10× slower |
| Chrome for rendering | `npx hyperframes doctor` (after step 2) | doctor says what to do |

## 2. Install (ask first)

1. `python -m venv .venv`
2. PyTorch — the command from pytorch.org for this OS; with an NVIDIA GPU pick the CUDA build (~2.5 GB), else CPU.
3. `.venv/Scripts/python -m pip install -r requirements.txt` (macOS/Linux: `.venv/bin/python`).
4. `npm install` (HyperFrames renderer, GSAP, Lucide icons).
5. `npx hyperframes doctor` — fix what it flags.
6. Tell the user: the Whisper `large-v3` model (~3 GB) downloads on the first transcription.

## 3. Fonts

Read `fonts` in the config. For each file missing from `font/`, tell the user where
the free default comes from and offer to download it (ask first):
- Farsi `Vazirmatn-Medium.ttf` — github.com/rastikerdar/vazirmatn (releases), OFL
- Latin `Poppins-SemiBold.ttf` and display `Poppins-Black.ttf` — fonts.google.com/specimen/Poppins, OFL

Any other font works: put the file in `font/` and change the path in the config.

## 4. Personal config

If `reel.config.json` does not exist, copy `reel.config.example.json` to it, then ask
in one message (skip what the arguments already answered):
- name and Instagram handle (for the follow card)
- spoken language of the videos (`fa`, `en`, `ar`, `de`, …)
- brand / product name (optional) and its colours — or a logo file to pick colours from
- mascot / character images (optional) → put them in `assets/mascot/`; a logo → `assets/brand/logo.png`
- where their B-roll clips are (default `assets/broll/`)

Write the answers into `reel.config.json` (keep every key the example has).

## 5. Mood board

Explain (in the user's language) that every edit starts from **their** mood board, so the
result looks like them and not like generic AI: `moodboard/default/` for their general
style, `moodboard/<video-name>/` per video. 5–15 screenshots of Reels they love, 1–3
short screen recordings of motion they like, their brand, and a `notes.md` with mood
words + likes + dislikes (docs/moodboard.md). Offer to look at it with them once it's filled.

## 6. Smoke test

- `.venv/Scripts/python -X utf8 -c "import sys; sys.path.insert(0,'tools'); import config, story_build, showcase_build; print('tools ok')"`
- If a video is already in `raw/`: `$PY tools/face_track.py <take>` should print a face position.

## 7. Hand-off

Print a short "how to use" with one example each, in the user's language:
`/reel-story <take> <topic, comment keyword, names>` and `/reel-split <take> …`.
```

## File: README.md
upstream: https://github.com/jenesyx/arta-edit-studio/blob/main/README.md

````text
<div align="center">

# 🎬 Arta Edit Studio

**Turn a talking-head phone video into a finished, animated Instagram Reel — by typing one command to Claude.**

**English** · [فارسی](README.fa.md)

Built by **Arta Bidkhori** · Instagram [@artabidkhori](https://www.instagram.com/artabidkhori) · MIT licensed, free forever

</div>

---

## Table of contents

1. [What is this?](#what-is-this)
2. [The two edit styles](#the-two-edit-styles)
3. [How it works](#how-it-works)
4. [Requirements](#requirements)
5. [Installation](#installation)
6. [Folder layout](#folder-layout)
7. [Configuration](#configuration-reelconfigjson)
8. [Preparing your media](#preparing-your-media)
9. [Your mood board (required)](#your-mood-board-required)
10. [Instagram safe zones](#instagram-safe-zones)
11. [Smooth motion](#smooth-motion)
12. [Using the commands](#using-the-commands)
13. [How to describe your video (prompting guide)](#how-to-describe-your-video-prompting-guide)
14. [Giving feedback after the first version](#giving-feedback-after-the-first-version)
15. [What you get](#what-you-get)
16. [Customising](#customising)
17. [Troubleshooting](#troubleshooting)
18. [Staying up to date](#staying-up-to-date)
19. [Credits and license](#credits-and-license)

---

## What is this?

Arta Edit Studio is a set of **skills (slash commands) for [Claude Code](https://claude.com/claude-code)**
plus the Python/Node tools they drive. You drop your video in a folder, type a command
like:

```
/reel-story my-video  I explain how my startup started. Already cut and coloured. Comment keyword: JOIN.
```

…and Claude acts as your video editor. It transcribes what you say, plans a beat for
every few seconds, picks moments from your own B-roll clips, writes the animations,
renders the final 1080×1920 video and checks it frame by frame before handing it over.

It was built for **Farsi (Persian)** creators first — right-to-left text, Persian digits,
Farsi fonts and Whisper settings are handled — but it works for any language Whisper
understands (English, Arabic, German, Turkish, …).

What it deliberately does **not** do:

- **No subtitles.** Both styles tell the story with graphics, not captions.
- **No cutting, colour grading or music.** It assumes you already edited your take
  (cuts, colour, music) in CapCut / Premiere / your phone. Your audio stays exactly as you mixed it.

---

## The two edit styles

| | `/reel-story` | `/reel-split` |
| --- | --- | --- |
| **Look** | You fill the whole screen. Cut-aways to your own clips, stickers and cards pop on your words, a mascot reacts, giant words appear **behind you**, the camera zooms, you shrink into a bubble over a map… | A cream panel with animated UI widgets fills the **top 45 %**. You sit underneath, still, the whole time. |
| **Best for** | Stories, announcements, startup / personal updates, invitations | "I found this tool", "5 tips", explainers, tutorials |
| **Uses** | Your B-roll folder, optional mascot + logo | Only your talking video |
| **Energy** | High — something changes every 2–4 s | Calm and clear — one idea per panel scene |
| **Ending** | Big comment keyword + follow card + logo lockup | Big comment keyword + follow card |

Both: no subtitles, no sound effects, your own audio untouched. And in both:

- **The look comes from your mood board**, not from a generic AI template.
- **Instagram's buttons, header and caption area stay clear** — checked automatically before rendering.
- **Every move is eased, curved and motion-blurred** — no robotic sliding text.

Signature tricks of `/reel-story` (all included as copyable examples):

- **Text behind you** — the tool cuts you out of your own video (AI person matte), so a giant word, your logo or a coloured world can sit *between* you and the wall.
- **Facecam bubble** — you shrink into a circle while a full-screen map or graphic takes over, then you grow back.
- **Mosaic zoom-out** — the camera pulls back until you are one tile in a wall of your own clips.
- **Fork card, comment typing, polaroid "right here", speed-ramped B-roll, exam stamps, coin rain, VHS rewind…** — 25 techniques in [`.claude/skills/reel-story/references/techniques.md`](.claude/skills/reel-story/references/techniques.md).

---

## How it works

```
raw/your-video.mp4
   │
   ├─ 1. Transcribe ........ Whisper large-v3 → every word with its exact time
   ├─ 2. Face tracking ..... OpenCV → where your face is (for zooms / framing)
   ├─ 3. B-roll catalogue .. contact sheets of your clips + a notes file Claude keeps
   ├─ 4. Mood board ........ your references → colours, pace, style → a style brief
   │                         that YOU approve before anything is designed
   ├─ 5. Beat plan ......... Claude pins a visual idea to a spoken word every few seconds
   ├─ 6. Assets ............ B-roll trimmed (HDR → SDR, mirrored selfies fixed),
   │                         mascot background removed, AI cut-out of you (matte)
   ├─ 7. Composition ....... Claude writes the animation as HTML + GSAP (HyperFrames)
   │                         with the motion library (curves, motion blur, keyframes)
   ├─ 8. Checks ............ lint, Instagram safe-zone scan, motion lint, snapshots
   │                         with zone guides, then a 12-second REAL test render
   ├─ 9. Render ............ HyperFrames renders 1080×1920 @ 30 fps
   └─ 10. Package + QA ..... your original audio put back, loudness check,
                             contact sheet of the result → final/your-video/
```

Claude is the editor (it reads, plans, writes the scenes and looks at the frames);
the tools in `tools/` do the heavy lifting. Everything runs **on your computer** —
no upload, no subscription other than Claude itself.

---

## Requirements

| What | Why | Notes |
| --- | --- | --- |
| **Claude Code** | the editor | Desktop app, terminal CLI or VS Code / JetBrains extension, with a Claude plan |
| **Python 3.10+** | transcription, face tracking, asset prep | |
| **Node.js 18+** | HyperFrames renderer, GSAP, icons | |
| **FFmpeg with `zscale`** | video processing, HDR → SDR | Windows: the "full" build from gyan.dev · macOS: `brew install ffmpeg` |
| **NVIDIA GPU** (recommended) | Whisper ~10× faster | Works on CPU too, just slower |
| **~10 GB free disk** | Whisper model (3 GB), PyTorch, renders | |
| **16 GB RAM** (32 GB comfortable) | rendering many video layers | Close games / Premiere while rendering |

Tested on **Windows 11 + RTX 3080 Ti**. macOS and Linux should work (every path is
cross-platform) but are not tested yet — reports welcome.

---

## Installation

### Easiest: let Claude do it

```bash
git clone https://github.com/Jenesyx/arta-edit-studio.git
cd arta-edit-studio
```

Open the folder in Claude Code and type:

```
/reel-setup
```

It checks your machine, asks before installing anything, downloads the free fonts,
and creates your personal `reel.config.json` by asking a few questions (name,
Instagram handle, language, brand colours).

### Manual

```bash
# 1. Python environment
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS / Linux:
source .venv/bin/activate

# 2. PyTorch — copy the right command from https://pytorch.org
#    (NVIDIA GPU → choose a CUDA build). Example for CUDA 12.4:
pip install torch --index-url https://download.pytorch.org/whl/cu124

# 3. The rest
pip install -r requirements.txt
npm install
npx hyperframes doctor        # checks the renderer and its Chrome

# 4. Your settings
cp reel.config.example.json reel.config.json   # then edit it (see below)
```

Then put the fonts in `font/` (see [Fonts](#fonts)).

---

## Folder layout

```
arta-edit-studio/
├─ .claude/skills/
│  ├─ reel-setup/        /reel-setup  — install + personal config
│  ├─ reel-story/        /reel-story  — full-frame motion edit (+ techniques catalogue)
│  └─ reel-split/        /reel-split  — 45 % panel edit
├─ tools/                the Python tools the skills run (+ motion.js)
├─ docs/                 mood board guide, style-brief template, mot
````
