# Download once. Let your coding assistant set it up.

Use **Codex in the ChatGPT desktop app** or **Claude Code**, with this complete folder open locally. Both use the same Make Video Skill tools. You choose your assistant and model; its plan, limits or credits are separate from this purchase.

After purchase and setup, rendering each video on your computer costs zero dollars. Save the finished MP4 files on your own computer. AI help and premium media can cost extra.

## 1. Put the toolkit in its own folder

Download your purchase from Gumroad, unzip it and move the complete `local-video-kit` folder to a place you want to keep, such as `Documents/MakeVideoSkill`. Do not run it inside the ZIP. Keep its hidden `.agents` and `.claude` folders, scripts, examples, assets and lockfile together. They are part of the product.

You only need to download and open the purchase once. The setup prompt below asks the assistant to handle free tool and dependency downloads for you where the computer permits. A sign-in, administrator password, installer agreement or system security prompt may still need your own action. Setup never includes an AI subscription or paid generation credits.

## 2. Open it in your assistant

### ChatGPT / Codex

1. Install the [official ChatGPT desktop app](https://learn.chatgpt.com/docs/quickstart), sign in and choose **Codex**. If your app has a different arrangement, follow that current official guide.
2. Open the extracted `local-video-kit` as the local project folder. Allow the file and terminal access needed for this folder.
3. Paste all of [the setup prompt](prompts/setup-all.txt) into a new chat there. You do not need to upload the ZIP to chat.
4. The included `.agents/skills/local-video-kit/SKILL.md` registers the folder's skill. If it is not listed, reopen the folder or restart the assistant; the prompt can still read the exact shared `SKILL.md` directly. In Codex CLI or the IDE extension, `$local-video-kit` can invoke it explicitly.

### Claude / Claude Code

1. Install the [official Claude desktop app and open its Code tab](https://code.claude.com/docs/en/desktop), or use the [official Claude Code terminal setup](https://code.claude.com/docs/en/setup). Sign in with your own eligible account.
2. Start a **local** session in the extracted `local-video-kit` folder. Allow this folder's files and the necessary local commands.
3. Paste all of [the same setup prompt](prompts/setup-all.txt).
4. The included `.claude/skills/local-video-kit/SKILL.md` registers the project skill. Type `/local-video-kit` to invoke it explicitly, or ask Claude to use this folder's `SKILL.md`. Reopen the session if the command does not appear.

Use the local Code/Codex workflow for the route described here. Uploading a ZIP to an ordinary browser ChatGPT or Claude chat is not the tested way to control the files and renderer on your own computer. Other agents with local files, a terminal and suitable permissions may run the same scripts, but are not individually verified. This is not a claim of support for every model, app or operating system.

## 3. Paste one prompt

Open [prompts/setup-all.txt](prompts/setup-all.txt), copy the complete text and paste it into the assistant. It tells the assistant to:

1. Inspect this computer and use the correct included adapter.
2. Install missing free prerequisites from reviewed official or well-known package sources when allowed.
3. Install the exact project dependencies and a separate Python environment for the nine styles.
4. Check FFmpeg captions, browser rendering, WebGL and the Python image/audio tools.
5. Make a real first video in both wide and vertical layouts, then collect the MP4s, captions and source in a new delivery folder.
6. Give you the finished local file links and any remaining action in plain language.

The supported automatic production setup is currently tested on **Apple Silicon macOS**. The full local speech workflow was exercised on a Mac with **32 GB memory**; this is a tested setup, not a proven minimum. Windows, Linux and Intel Mac buyers should obtain platform-specific confirmation before purchasing. A vendor's support for its own app on an operating system does not establish that this complete toolkit was tested there. There is no promised installation time.

## 4. What the setup installs

| Item | Why it is used | How it is obtained |
| --- | --- | --- |
| Node.js 20.9+ and npm | Runs the toolkit | Official Node distribution; the assistant can use an existing trusted package manager. |
| FFmpeg + FFprobe, including libx264 and the `ass` caption filter | Makes MP4s, checks media, draws captions | A trusted FFmpeg build; the assistant can use an existing Homebrew installation on Mac. |
| Python 3.11+ | Finishes the three character/documentary styles | Python 3.12 is the tested choice. |
| Pinned npm dependencies | Drawing, fonts and local browser automation | `package-lock.json`, installed inside the toolkit. |
| Pinned NumPy + Pillow | Sound and image finishing | Toolkit-local `.mvs-production/venv`, from `style-lab/expanded/requirements-production.txt`. |
| Chrome or Chromium | Renders the native animated scenes and film text | Uses an installed compatible browser; otherwise downloads the Chromium revision belonging to the pinned Playwright package into `.local-tools/browsers`. |

`node scripts/assistant-setup.cjs install` does the project dependency, Python-environment and optional Chromium steps after the base tools are available. It does not buy anything, change security settings or install a global model. To choose the toolkit-local browser even when Chrome is already installed, use `node scripts/assistant-setup.cjs install --managed-browser`. It saves its local tool paths inside `.mvs-production/environment.json`; keep that private to this computer.

Use `node scripts/assistant-setup.cjs check` for the expanded read-only check. Run a style command with the saved environment like this:

```sh
node scripts/assistant-setup.cjs run -- scripts/styles.cjs list
node scripts/assistant-setup.cjs run -- scripts/styles.cjs init --style plush --out projects/my-plush
node scripts/assistant-setup.cjs run -- scripts/styles.cjs render projects/my-plush --out outputs/my-plush --format both
```

Existing project and export folders are protected. Choose new names when repeating an exercise.

## 5. Make your own video

Open [prompts/new-video.txt](prompts/new-video.txt), fill in your topic and paste it into the assistant. The assistant should show the hook, short script and shot plan before expensive generation or a long final render. Start with a supplied style or one of the three starter projects: explainer, product demo or recording edit.

Use [prompts/change-video.txt](prompts/change-video.txt) for changes. Describe what you want in everyday language. A visual text or colour change does not rewrite recorded speech. New spoken words require new audio, word timings and matching scenes. Supplied examples include finished narration; realistic external UGC presenters and private speaker profiles are not supplied free generation services.

For **new free local synthetic narration**, use [prompts/new-local-voice.txt](prompts/new-local-voice.txt). It guides the assistant through the separate Apple Silicon MLX environment and pinned model downloads. The voice, alignment and planning models total roughly **6.7 GB**, plus packages, caches and video space. They are not needed for the supplied examples and are not hidden inside the purchase ZIP. Optional premium voices or generated footage use the buyer's chosen provider and can cost extra.

## 6. Finish and keep your files

Watch the wide and vertical MP4s from beginning to end with sound. Check speech, captions, hands and faces, crops, transitions and the last frame at phone size. Then use the included delivery helper and [export checklist](references/export-checklist.md). The output includes videos, captions, covers, editable project source, rights records and checksums. Keep the full toolkit separately so you can rebuild the project later.

If something fails, ask the assistant to fix the reported stage and preserve earlier work. Use [setup-troubleshooting.md](references/setup-troubleshooting.md) for manual alternatives and a privacy-conscious support report. Do not send passwords, API keys, raw environment dumps or your private footage when asking for setup help.

Official app and skill-location details were checked on 31 October 2026. See [assistant compatibility](references/assistant-compatibility.md) for sources and the boundary between documented app support and recorded toolkit tests.