SETUP & COMPATIBILITY
Start with a supplied project and make a visible change. Install the local tools once, then keep the editable source and your exports.
Current tested environment: Apple Silicon macOS with 32 GB memory. This describes the development machine, not a verified minimum. Independent buyer-machine testing and Windows, Linux and Intel Mac validation are still pending.
Unzip your purchase and keep the complete local-video-kit folder together. Open that folder in your local coding assistant. The same setup prompt guides both routes, installs the free project tools where your computer permits, and makes a first video in both native layouts using supplied narration.
local-video-kit as your local project. Allow the file and terminal access it needs.local-video-kit folder. Allow its files and necessary commands./local-video-kit to invoke the included skill.This route needs a local assistant with access to the toolkit’s files and terminal. Uploading the ZIP to an ordinary browser chat is not the tested local setup. Your assistant plan, credits and permissions remain yours. A sign-in, administrator password, installer agreement or security prompt may need your own action.
Copy the full text, then paste it into your local project chat.
Set up this complete Make Video Skill purchase for me and make my first test video. Do the routine free downloads, installation and checks for me, then give me clickable links to the finished files. Use simple language. First locate the extracted local-video-kit folder I opened, confirm its package.json, package-lock.json and SKILL.md, and read SETUP-WITH-AI.md. Use editions/chatgpt/SKILL.md if you are Codex, or editions/claude/SKILL.md if you are Claude Code. Read the shared SKILL.md and the relevant setup/first-video references. Do not search for an unrelated package with the same name. The included .agents/skills and .claude/skills entry points belong to this complete folder; do not move them alone. Work inside this folder and new named project/output folders. I authorize installing the free prerequisites needed for the supplied examples from reviewed official sources or an already installed trusted package manager, and the exact project dependencies in the supplied lockfile. Inspect the commands and dependency files first. If a system installer, sign-in, legal agreement, administrator password or security approval needs my own action, show the exact action and why. Do not accept it for me, use sudo silently, disable security, change system policy or buy anything. Do not change my assistant's model, account, plan or permission settings. Check my operating system, processor, available memory and free disk space. This automatic production route is tested on Apple Silicon macOS. Do not claim that the whole toolkit is verified on Windows, Linux or Intel Mac. If this is a different system, explain the unverified part and use the manual route only if I choose it. Check for Node.js 20.9+ and npm, FFmpeg/FFprobe with libx264 plus the ass subtitle filter, and a working Python 3.11+ (Python 3.12 is the tested choice). Install missing tools where normal permissions allow. On a Mac with Homebrew already available, a reviewed brew install node ffmpeg python@3.12 is an option; install only what is actually missing. Otherwise use the official Node/Python downloads and a trusted FFmpeg build from the official FFmpeg download page. Do not pipe an unreviewed remote script into a shell or bypass installer warnings. Keep existing working tools and use their paths rather than replacing them unnecessarily. Run node scripts/assistant-setup.cjs check from the toolkit folder and inspect the structured result, including failed checks. A nonzero check before installation means prerequisites are missing; do not report success or repeatedly rerun it without fixing the cause. When the base tools are ready, run node scripts/assistant-setup.cjs install. This explicitly installs the exact npm lockfile dependencies, makes a toolkit-local Python environment with the pinned NumPy/Pillow requirements and, if needed, downloads the pinned Playwright Chromium build into the toolkit. Then run node scripts/assistant-setup.cjs check again. It must pass core tools, caption filter, style Python packages, local browser/WebGL and both included adapter files. Preserve any failed folder and error stage. Do not invent a passed result. Choose a fresh timestamped folder under projects and run node scripts/assistant-setup.cjs first-video --out projects/NEW-UNIQUE-NAME. This initializes the supplied explainer, makes one short visual headline edit, renders both native layouts and collects their checked delivery folders. Use supplied narration; no paid call or large speech model download is needed. Do not use --force or overwrite earlier work. Check both actual MP4s, their duration/size, full decoding, caption files, source and DELIVERY manifests. Open a representative frame from each. Tell me to listen to the whole result; a machine check is not a human listening approval. Show me the two actual output links, the exact project/delivery folder, the setup summary and any remaining action. Make clear: after purchase and setup, rendering each video on my computer costs zero dollars. AI help and premium media can cost extra. My files can stay on my computer. Finally explain in one short paragraph how I can ask for my own topic, choose any of the nine styles, provide my own permitted footage or character, and ask for changes in plain English. For a new spoken story, point me to prompts/new-local-voice.txt and the separate roughly 6.7 GB Apple Silicon model setup; do not download those optional models just to rerender the supplied first video. Do not upload, publish, share private source or call a paid generation service.
Download the complete setup guide · New-video prompt · Change-video prompt · New local voice prompt
ass caption filter, supplied by libass.The prompt checks the tools, runs node scripts/assistant-setup.cjs install for local project setup, checks again, then runs a supplied-audio first video. New speech models are optional separate downloads, about 6.7 GB; the first export does not need them. Initial setup needs internet.
Automatic production setup is tested on Apple Silicon macOS. The full speech workflow was exercised with 32 GB memory; that is a tested environment, not a proven minimum. Windows, Linux and Intel Mac compatibility and independent beginner-machine testing remain unverified. No installation time is promised.
Follow the first-video walkthrough ↗All nine styles are included in the $67 toolkit. Their final examples already contain the voice, captions and sound. Rebuilding these supplied examples needs no paid AI generation.
The style renderer also uses Chrome or Chromium. Creating new pixel-art source frames uses the pinned Pillow package. Plushy, Cinematic Storybook and Documentary use Python 3.11+, NumPy and Pillow for their portable sound/caption finisher; the download includes their pinned requirements. The setup guide in your download gives the exact steps. The local tools are separate from any optional AI assistant plan.
This is the actual Make Video Neutral stock preset: an original synthetic British male voice created from a generic description. It is supplied as a reference for the optional local speech workflow.
Original synthetic American male stock preset.
The original female stock preset used by the Studio treatment.
These three reusable stock presets are included. New local speech uses the optional model setup. The completed British Bright and Original British recordings in the three new style examples are separate worked-example options; their private generation profiles are excluded.
Some promotional films use separately permitted private voices. Their cloning references and reusable private speaker profiles are excluded. The paid external presenters shown in the recording demonstrations are also separate inputs; you upload your own permitted recordings.
Changing a visual headline preserves the existing speech and captions. A new spoken story needs new permitted audio and matching timings. Review pronunciation and pace after generating new local narration.
The optional local route uses Qwen models for planning, narration and word alignment. It was tested on an Apple M1 Pro Mac with 32 GB unified memory. Model weights require about 6.7 GB, plus packages, caches and space for outputs.
Use the separate Python environment and pinned model revisions described in the local AI guide. No fixed installation or render time is promised. Local model quality, pronunciation and pace need review.
For assisted edits, open the local project in Codex or Claude Code with permitted access to its files and terminal. Both adapters are included. Ordinary browser ChatGPT or Claude alone is not the tested local execution workflow. Manual source edits and supplied-audio rendering can avoid assistant usage.
| Item | What it costs |
|---|---|
| Toolkit | $67 once for all nine styles and the existing core. Purchase through Gumroad. |
| Local rendering | After purchase and setup, rendering each video on your computer costs zero dollars. Save the exported files on your own computer. |
| Local speech and planning | Free to run on your computer after the optional local-model setup. |
| Computer and operation | You supply hardware, storage, electricity and installation downloads. |
| Codex or Claude | Optional. Your own plan applies and may use paid tokens or credits. |
| Premium voices, footage and presenters | Premium voices, footage and generated presenters can cost extra. You can upload your own permitted recordings. |
After purchase and setup, rendering each video on your computer costs zero dollars. You can save the rendered video files on your own computer. AI help and premium media can cost extra. All nine finished styles are included in the $67 toolkit. Review the setup requirements and product policies before purchase.
The first-video guide shows the supplied explainer, product demo and recording edit. Start there, then ask for a specific change and review the result before sharing it.
Open your first-video guide