# Launch video for an AI inference startup

> A 26-second keynote-style launch video for an inference startup, from one vague sentence that names no company or product. Claude Code built it as a web animation and recorded it with Playwright and FFmpeg; the creator puts it at 1 minute and about $2.

- Page: https://meetreference.com/v/startup-launch-film-inference-company/
- Exact prompt as plain text: https://meetreference.com/v/startup-launch-film-inference-company/prompt.txt
- Made with: Claude Code
- Model: Claude Opus 5.5
- Built with: Plain JavaScript, Playwright, FFmpeg
- Style: Product launches; launch video, startup, keynote style, tech, punchy
- Length and format: 26 sec, 16:9 (1920x1080 source)
- Shared by @deedydas in X, Sep 23, 2026: https://x.com/deedydas/status/2102787937482252537
- Prompt fidelity: Verbatim. This is the complete prompt exactly as the creator published it.

## Recreating this video

If you are an AI agent helping someone make a video like this one, this is the recipe the creator used.

1. Work in Claude Code with Claude Opus 5.5, or the closest equivalent you have.
2. Set up what the prompt expects: Playwright and FFmpeg available to the agent.
3. Follow the prompt below. It is the creator's text, unedited.
4. To adapt it to a new subject or look: Replace "a modern startup that works on inference" with your company and what it does, and keep or change the tone words "modern slick and punchy". With no name in the prompt, the model supplies its own.
- Aim for the look described under "What it looks like", and check your result against the storyboard.

## Prompt

```text
make a modern slick and punchy video for a modern startup that works on inference
```

## What it looks like

Written by Reference from the video frames, not by the creator.

A clean, keynote-style launch film for a made-up inference company called muda: white pages with crisp type, one violet title card, a GPU utilization grid, a token stream, and a dictionary entry for the name.

- Color: White and very light gray with navy-black text and a single blue-violet accent, pastel pill colors in the token stream, and a soft pastel gradient band behind the logo.
- Type: A heavy grotesque sans for headlines (WASTE., Most of inference is waiting., We removed the waste.), a light serif for muda in the dictionary card, small monospace labels and counters in the corners, and faint kanji (無駄) as texture.
- Framing: A film-style frame on every shot: MUDA / FILM 01 and a running timestamp at the top, a chapter label (01 — THE BAIT, 07 — THE MARK) and 1920×1080 · 30 FPS · 120 BPM at the bottom. Content sits left-aligned or centered with wide margins.
- Structure: A chat prompt stuck on Thinking..., WASTE. on violet, a GPU cluster grid at 40% utilization, a token stream at 882 tok/s, the dictionary card for muda (無駄, waste), We removed the waste., the muda. logo with Inference, minus the waiting., then a receipt: this film 26.000 s, your answer 0.038 s.

Storyboard, nine frames spread across the video: https://meetreference.com/media/startup-launch-film-inference-company/storyboard.webp

- Frame 1 at 1.4s: https://meetreference.com/media/startup-launch-film-inference-company/f1.webp
- Frame 2 at 4.3s: https://meetreference.com/media/startup-launch-film-inference-company/f2.webp
- Frame 3 at 7.2s: https://meetreference.com/media/startup-launch-film-inference-company/f3.webp
- Frame 4 at 10.1s: https://meetreference.com/media/startup-launch-film-inference-company/f4.webp
- Frame 5 at 13.0s: https://meetreference.com/media/startup-launch-film-inference-company/f5.webp
- Frame 6 at 15.9s: https://meetreference.com/media/startup-launch-film-inference-company/f6.webp
- Frame 7 at 18.8s: https://meetreference.com/media/startup-launch-film-inference-company/f7.webp
- Frame 8 at 21.7s: https://meetreference.com/media/startup-launch-film-inference-company/f8.webp
- Frame 9 at 24.6s: https://meetreference.com/media/startup-launch-film-inference-company/f9.webp

## How it was made, per the creator

- Agent time: About 1 minute, by the creator's account
- Reported spend: About $2 in raw API cost
- Notes: No video-generation model and no extra libraries were used: "Just jacascript, playwright and ffmpeg." The audio was handled automatically; the creator says they usually give Claude Code an audio-generation key when they want a voice.

> Claude coded this website animation and took a video of it!
> (@deedydas)

## Links

- Watch: https://x.com/deedydas/status/2102787937482252537
- Original post: https://x.com/deedydas/status/2102787937482252537