# Kinetic-type launch film for an AI startup

> A 32-second launch film for a made-up inference cloud called Infera, with big headlines on alternating color fields, a request-router diagram, and a code snippet. LaunchVideo's agent had Opus 5.5 write it as one HTML page, then rendered it frame by frame.

- Page: https://meetreference.com/v/kinetic-type-launch-video-up-inference/
- Exact prompt as plain text: https://meetreference.com/v/kinetic-type-launch-video-up-inference/prompt.txt
- Made with: OpenComputer agent (LaunchVideo)
- Model: Claude Opus 5.5
- Built with: HTML and CSS, Playwright, FFmpeg
- Style: Product launches; launch video, kinetic typography, startup, flat color, diagram, editorial frame
- Length and format: 32 sec, 16:9 (1920x1080 source)
- Shared by diggerhq in Website, Sep 24, 2026: https://launchvideo.io
- 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 OpenComputer agent (LaunchVideo) with Claude Opus 5.5, or the closest equivalent you have.
2. Set up what the prompt expects: Use launchvideo.io, or deploy the agent to your own OpenComputer account (npx opencomputer template deploy https://github.com/diggerhq/shipvideo); In another agent, add the director's public system prompt from the repo for the same look; the one line alone leaves the style open.
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 product, or give LaunchVideo a URL instead and it reads the page for copy, colors, and fonts. Keep "modern slick and punchy" for the tone.
- 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 nine-beat launch film for a made-up inference cloud: big two-line headlines on black, acid-lime, and violet fields, a few simple mechanism graphics, and an end card for Infera.

- Color: Near-black and acid lime with one violet beat, white and lime type, lime interface details, and dark code boxes.
- Type: A wide geometric sans in a heavy weight for two-line headlines, the second line in lime, and small monospace labels for the frame (INFERA / LAUNCH, BEAT 01/09, a running timecode, INFERENCE CLOUD), section tags (01 — ROUTE), and the code block.
- Framing: Headlines set left-aligned in the upper left, with one diagram, chart, or code box below or beside them. A thin editorial frame of corner labels stays fixed, with a progress hairline along the bottom.
- Structure: Your model is ready., Your users are waiting. over a progress bar, GPUs idle. Bills climbing. beside a GPU grid and a rising line, One endpoint. Any model. with a request, router, GPU pool, response diagram, Tokens, streaming., Scales up. Scales to zero. over a bar chart, Ship it in one call. with a curl command, and the Infera end card with infera.dev.

Storyboard, nine frames spread across the video: https://meetreference.com/media/kinetic-type-launch-video-up-inference/storyboard.webp

- Frame 1 at 1.8s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f1.webp
- Frame 2 at 5.3s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f2.webp
- Frame 3 at 8.9s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f3.webp
- Frame 4 at 12.4s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f4.webp
- Frame 5 at 16.0s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f5.webp
- Frame 6 at 19.6s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f6.webp
- Frame 7 at 23.1s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f7.webp
- Frame 8 at 26.7s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f8.webp
- Frame 9 at 30.2s: https://meetreference.com/media/kinetic-type-launch-video-up-inference/f9.webp

## How it was made, per the creator

- One-shot: yes, a single prompt with no follow-ups
- Agent time: About four minutes per video, per the site
- Reported spend: Roughly 100k tokens per video, per the site
- Notes: The site calls each example "a single run: a URL or a prompt in, an MP4 out. No edits." The agent runs on OpenComputer with Claude Opus 5.5 and three tools: web_fetch, check_scene (loads the film under a virtual clock and reports errors and visible text), and render_video (Playwright frames into FFmpeg at 1920x1080). Its system prompt, public in the repo, sets the style: 20 to 40 seconds in 6 to 10 beats, showing the mechanism, one or two bold background colors, and a thin editorial frame with a counting timecode. The typed prompt is @deedydas's, reused word for word.

## Links

- Watch: https://launchvideo.io
- Original post: https://launchvideo.io
- Hacker News discussion: https://news.ycombinator.com/item?id=49836374
- Source code and agent prompt on GitHub: https://github.com/diggerhq/shipvideo
- The first video from this prompt, by @deedydas: https://x.com/deedydas/status/2102787937482252537