How to make a product launch video with Claude or ChatGPT.
What do you need to make a launch video with AI?
Four things, whether you build it yourself or use a tool:
- A coding agent with a strong model. Claude Code with Claude Opus 5.5, or OpenAI's Codex. The model writes the film as code.
- A renderer. Software that runs that code for every frame and writes an MP4. HyperFrames (HTML and GSAP, open source), Remotion (React) and fframes (Rust and SVG) all work this way.
- Music and sound. A real song cut to the edit, and small sound effects on the presses and landings. 89% of the viral films have sound.
- Checks. A way to measure the render: how much of it stands still, how many hard cuts, how loud, whether sounds land on motion. Without this, "looks fine" is the only bar, and it's a low one.
What are the ways to run it?
| You have | How it runs | What you pay |
|---|---|---|
| Claude Pro or Max | Inside Claude Code, which comes with the plan. TeaserKit installs its skill there and hands you one line to paste. | Your plan's normal limits |
| ChatGPT Plus or Pro | Sign in with ChatGPT inside TeaserKit; it runs on Codex. | Your plan's normal limits |
| An API key | Claude, GPT, Gemini, Grok, Kimi, GLM, DeepSeek, Qwen or OpenRouter. | What each video uses, with a spending cap ($15 default) |
Anthropic doesn't let other apps sign in with a Claude plan, which is why the Claude-plan route runs inside Claude Code itself.
How do you make one, step by step?
- Write the message first, word for word, in under 15 words. "We're launching v2 today. Invoices send themselves." The film exists to land that line.
- Give it real material. Your website link, your logo, two or three screenshots, one real number. The model is very good at layout and bad at guessing facts.
- Name a look, don't describe a vibe. "Apple product-page micro-interaction" or "movie-trailer cold open" beats "make it premium". TeaserKit has 11 house styles for this.
- Let it plan before it animates: three concepts, the strongest one, a storyboard cut to the beat of the song.
- Check the render, not the code. Stills of every key moment, then a draft, then the checks. Fix the worst failure, re-check.
- Revise in plain words. "Make the ending punchier." "Use our green instead of orange." "Make the music calmer, no bass."
What do you get at the end?
- A 16:9 MP4 at 1920×1080 and 60 fps for X, LinkedIn and your site.
- A 9:16 vertical version (1080×1920) for Reels, TikTok and Shorts.
- A contact sheet of every frame with review notes, and the project files, all on your own computer.
How do you make it not look AI-made?
The difference between a "mid" first render and a film people share is mostly measurable. From the 74 viral films we measured: about 30 seconds long, about 2 hard cuts per 30 seconds, under 10% of the time with nothing moving, mastered around −14 LUFS. And from our own films: sound placed on the frame where things actually land, not where a storyboard guessed. The full method, with code and commands, is in How to turn any coding agent into a motion studio.
Want it done for you?
TeaserKit runs this whole workflow on the Claude or ChatGPT plan you already pay for: concept, storyboard, a real song cut to the beat, the checks and the critic, and a 60 fps render. One-time purchase.
Questions people ask
Can Claude make videos?
Not directly as a video file. Claude writes code that draws each frame (HTML, CSS and animation), and a renderer turns that code into an MP4. That is how the Opus 5.5 launch films on X were made: every frame rendered from code, no After Effects and no video model.
Can ChatGPT make a launch video?
Yes, the same way: OpenAI's coding agent (Codex) writes the video as code and a renderer turns it into an MP4. TeaserKit can run on a ChatGPT Plus or Pro sign-in.
Do I need to know how to code?
No, if you use a tool that wraps the workflow: you describe the video in plain words and ask for changes the same way. Without a tool you will be reading and running code, renderer commands and FFmpeg.
How long does it take?
A 30-second video usually takes 20 to 45 minutes with TeaserKit, depending on the AI model. Changing one scene is quicker, and swapping the music takes seconds.
How much does it cost?
On a Claude Pro/Max or ChatGPT Plus/Pro plan, videos use your plan's normal limits. With an API key you pay the AI company for what each video uses; TeaserKit stops each video at a spending cap you choose ($15 by default). TeaserKit itself is a one-time purchase.
Why do most AI launch videos look the same?
Without a reference and rules, models fall back to the same default: a centred title on a gradient with everything fading in. Naming a style, giving real product material and checking the render against measured targets is what changes the result.