How to Make a Product Launch Video With AI, Music Included

A 30-second product launch film with its own music, drawn entirely in code, from one sentence typed into Claude Code. That is what huashu-motion does, a free skill an indie developer called Huashu put on GitHub on 18 October 2026. By 27 October it had 1,775 stars.

You tell the agent "make a 30-second launch film for my product" and give it your website. It reads the product, shows you three directions, writes the film, composes a score for it and opens an editor in your browser. If one part is off, you press C, click on it and type what you want changed, and only that part changes.

This guide covers the whole job: what to install, how to write a brief that gets a good first cut, how to direct the edits, how to export for ads and social, and what to check before you publish.

What huashu-motion is

It is a skill: a folder of instructions and tools an AI agent loads when the job calls for it. It follows the Agent Skills format, which Anthropic first developed and released as an open standard, so any agent that supports skills can install it. The repo's README lists Claude Code, Codex, WorkBuddy, Grok Bot, Muse and Dots.

There is no video generation model involved. Each film is a single web page, and the README puts it this way: "every frame is a function of the code and the time." The motion is drawn with browser tools like GSAP, Canvas, Three.js and SVG, and the skill's own render engine turns that page into a video file, running several Chrome windows at once to do it.

The music is made the same way. For launch films the agent writes a score in ABC notation, a plain-text form of sheet music, with marks that land on the scene changes. A sampled grand piano, strings and drums play it, and one of 23 sound effects lands on the frame where each animation lands. Everything is generated on your computer.

Because the film is code, the parts a marketer cares about stay editable. Titles, numbers, colours and hold times show up as fields you can change yourself, without asking the agent.

Huashu (花叔), who publishes as alchaincyf on GitHub and @AlchainHust on X, says he started the project after Anthropic released Claude Motion on 8 October 2026, which launched as a beta for Team and Enterprise plans only. The skill's own 53-second launch film was made with the skill.

Before you start

• An agent to run it. Claude Code needs a paid Claude plan: Pro is $20 a month, or $17 a month billed yearly (Claude pricing). Codex is included in ChatGPT plans from Free upward, with usage limits.
• Node.js 22 or newer from nodejs.org. The current long-term support release is Node 24.
• Google Chrome, Chromium or Edge. Stills, checks and export run in the browser you already have, so nothing extra is downloaded for rendering.
• FFmpeg, the free tool that encodes the video and soundtrack. The FFmpeg project only ships source code, so you install a ready-made build (steps below).
• A Mac if you want the tested path. The creator says it is tested on macOS only so far. Windows and Linux may need more setup.
• Your website, and ideally a short note on who the product is for and what a viewer must understand.

Stage 1: Install Node.js, a browser and FFmpeg

1. Install Node.js 22 or newer from nodejs.org with the default options. In a new terminal, node --version should print v22 or higher.
2. Make sure Chrome, Chromium or Edge is installed. Edge ships with Windows, so Windows users already have one.
3. Install FFmpeg. On a Mac with Homebrew, run brew install ffmpeg. On Windows, run winget install Gyan.FFmpeg, or download a build from gyan.dev or BtbN, both linked from FFmpeg's download page.
4. Open a new terminal and run ffmpeg -version. A version banner means it is on your PATH. If the command is not found, close and reopen the terminal, or ask Claude Code to fix the PATH for you.

You can skip all of this and let the agent do it. Open Claude Code and say "install Node 22, Chrome and FFmpeg for me if they are missing."

Stage 2: Install the skill

Open Claude Code in an empty folder for the project and paste the first line of the prompt below. The README offers the same install as one terminal command, if you prefer that.

CopyInstall this skill and run the demo: https://github.com/alchaincyf/huashu-motion

Then make a 30-second launch video for my product. My website is [your site].

Copynpx skills add alchaincyf/huashu-motion

Prefer a one-click copy? The same prompt is on our free prompt page.

1. Claude installs the skill and runs the demo film. If your session is in manual mode (Shift+Tab switches modes), approve the commands it asks to run.
2. If anything is missing, ask the agent to run om doctor, the skill's environment check. It lists what is not installed. Say "install it for me" and it will.
3. Optional but worth it: ask it to run om samples build. This downloads about 55 MB of CC0 and CC-BY instrument samples once. Without them, the music plays on synth sounds.
4. In Codex, the sandbox blocks the local preview and the browser by default. Approve them when Codex asks, or you will not see the preview.

Stage 3: Write a brief that gets a good first cut

The short prompt works. The creator's own notes on his sample films are honest about the rest: one line does make a film, but to get to his level, either the first message carries the brief or someone keeps directing. A fuller brief saves you rounds of notes. Here is the shape we suggest:

CopyUse the huashu-motion skill. Make a 30-second launch film for my product.
Website: [your site]
Who it is for: [one line about the buyer]
A viewer must come away knowing:
1. [what it is]
2. [the one result it gets them]
3. [how to get it: price, offer or link]
Format: 16:9, English, end on [your URL or call to action]
Show me three directions as key frames before you build.

Why each line matters:

• The website gives the skill real material. Its om grab command pulls page screenshots, your logo, images, your colours and your typefaces into the project, so the film is built from your own brand material.
• The three must-knows set the story. In one of the creator's briefs, he listed three things every version had to get across and let the agent decide how.
• Format avoids a re-layout later. Say 16:9 for YouTube and your site, or ask for vertical from the start if the film is mainly for Reels and TikTok.

The skill brings its own director rules, measured from real launch films: a hook within 3 seconds, a burst of energy in the middle and a beat of silence before the biggest moment. You do not need to ask for those.

Stage 4: Pick a direction and let it build

1. The agent reads your product and shows three directions as key frames. Pick one, or combine them. On one of his films the creator simply replied "I'd like a and b combined."
2. It writes a rhythm sheet, builds the film and scores it. This is the long part, and you can leave it running.
3. Before it shows you anything, it runs its own checks: om check for broken scripts, motion rules and text that is cut off or covered, and om qa for single-frame pops, dead time and the motion curve.
4. It opens the editor in your browser with a live preview and a strip of scenes. Play it through once with the sound on before you change anything.

Rendering is fast once the film exists. The creator reports a 30-second film renders in about 12 to 15 seconds on his 14-core Mac, or about a minute to a minute and a half with motion blur.

Stage 5: Fix parts by pointing at them

This is where the skill saves the most time. You do not describe the problem from memory. You point at it.

1. In the editor, pause on the frame that is off and press C.
2. Click the spot in the frame that is wrong, such as a headline or an icon.
3. Type the change in plain words and send it.
4. Your comment carries the time, the scene, the object under your cursor and its line in the source code. The agent looks at that frame and a close-up of the spot first, changes only that, replies in one line and reloads the preview at the moment you were watching.
5. Every version is kept, and ⌘Z undoes a change. For a headline, a number, a colour or a hold time, skip the comment and edit the field directly.

Notes that work are short and specific about what is wrong. The creator's own notes on his films are a good model: "the first frame of the opening is poor; better to focus it on the typing", "drop this effect", "the ending is a bit long", "the big moment could start earlier." The problems he flagged most (a weak opening, sections that look like slides, flat rhythm, repetitive material) are now checks the skill runs on itself before it shows you a film.

Stage 6: Listen, then export

1. Listen to the whole film with good headphones. The README is direct: "It can't hear." The agent checks music through the score, an energy curve and a spectrogram, and "the last listen is yours."
2. Check every claim, price and number on screen against your website. The film is only as accurate as what the agent read.
3. Say "export it as an MP4." Other options: MOV with a transparent background, WebM, GIF, or a share page with a player.
4. For social, say "make it vertical." You get a 1080 by 1920 copy, with the re-layout waiting in the editor's inbox for review.
5. After export, ask for om share. It makes a GIF under 10 MB (the opening plus the end frame) and a share.md file with draft post text.

Films of 4 seconds or more end with a small "Made with huashu-motion" for their last 1.2 seconds, in a free corner, in the film's own typeface and colour. Ask the agent to switch it off if you do not want it.

Other jobs the same skill does

Launch films are the headline, but the README lists other requests, and each one is a prompt you can try:

It also ships 12 starter templates, from keynote and finance chart to kinetic type, whiteboard and objects in a 3D studio. Narration can use a macOS system voice with no setup, or Kokoro voices, which need uv, Python 3.12 and a model of about 325 MB. Word timing for your own recording needs whisper-cli (brew install whisper-cpp on a Mac).

What it costs

You do not need the most expensive model. Huashu made a 22-second film with Claude Code running on DeepSeek V4.1 Flash, gave three rounds of notes, and says it took under half an hour and cost 1.95 yuan at the time. He mostly uses Opus 5.5 himself and notes that taste differs between models.

Troubleshooting

Before you run it as an ad

• Bring your own footage if you need people. The skill does not generate people or live footage. Everything is drawn in code, and if you give it footage, it composes and packages that.
• Check the licences for commercial use. The code is MIT, but bundled pieces keep their own licences: GSAP under the GSAP Standard License and the fonts under SIL OFL 1.1, listed in the repo's THIRD_PARTY.md. Huashu's cartoon character is excluded from the MIT licence and is for the project's demos only.
• Use your own brand. The Grok Bot and Claude Duet films on the releases page are unofficial fan films. Point the skill at your own website so the film is about your product.

Other ways to get a launch video

If your company is on a Claude Team or Enterprise plan, Claude Motion is Anthropic's built-in version. Anthropic's 8 October 2026 announcement says Claude builds an animation from your prompt that "plays like a short video," you adjust it in an editor or by asking, and you download it as an MP4. It launched in beta for Team and Enterprise only.

If you need real people in the film, a product being used or a founder talking, a code-drawn film will not cover it, and you are back to filming. You can use both: a motion film for the announcement and a filmed clip wherever a face matters. If you would rather have someone handle the whole launch creative, our creative production team does that work for businesses.

Frequently asked questions

Can AI make a product launch video for free?

The huashu-motion skill is free and open source under an MIT licence, and it generates the film and music on your computer with no paid API. You still need an agent to run it: Claude Code needs a paid Claude plan, while Codex is included in ChatGPT plans from Free upward with usage limits.

What is huashu-motion?

It is a skill for Claude Code, Codex and other agents that support the Agent Skills format. You ask for a launch film for your product, pick one of three directions, and it writes the film as code, composes the music and opens an editor where you click on any part to change it. It was made by Huashu (alchaincyf on GitHub).

Does huashu-motion make the music too?

Yes. The agent writes a score in ABC notation that lines up with the scenes, and sampled piano, strings and drums play it after a one-time samples download. Without the samples it uses synth sounds. Sound effects land on the frames where the animation lands.

Does huashu-motion work on Windows?

The creator says it has only been tested on macOS so far. It needs Node 22 or newer, Chrome, Chromium or Edge, and FFmpeg, so Windows may take more setup. Ask Claude Code to install anything missing, then have it run om doctor to see what is still needed.

Can I use the video in my ads?

Yes, the export is a normal MP4 you can run as an ad, post on social or put on your website. The skill's code is MIT licensed, while bundled parts like GSAP and the fonts keep their own licences listed in THIRD_PARTY.md, so check those for commercial use. Ask the agent to remove the small "Made with huashu-motion" credit if you do not want it.

Can it put real people or my own footage in a launch film?

It does not generate people or live footage, because everything is drawn in code. If you give it your own footage or a voice recording, it can compose the film around it, with word timing and subtitles for narration.

Sources

• huashu-motion on GitHub (README and English README)
• huashu-motion releases (demo films)
• Agent Skills overview
• Anthropic: Claude Dashboards and Claude Motion (8 October 2026)
• Claude pricing and Codex pricing
• Node.js releases
• FFmpeg downloads