TakeOne: the launch demo video your agent films itself

The problem

A decent demo video costs $1,600+ and 20 days from an agency, so founders launch with rough screen recordings that make the product look worse than it is. Avatar-video tools (Synthesia, HeyGen) solve a different problem: they show a talking head, not your product being used.

The idea

Why now

Open computer-use stacks just went mainstream — Alibaba OpenGCUA (October 16) and H Company Holo4 publish open weights and full stacks for agents that drive real UIs — while Kling 4.0 (October 16) brings professional-grade video generation. Agent-filmed product footage, which needed a bespoke robotics-style rig a year ago, is now composable from open parts.

What it combines

Alibaba OpenGCUA / H Company Holo4 (open computer-use agents that can drive a real product) + Kling 4.0 (professional video generation) + deterministic capture pipelines (Remotion-style). Avatar-video tools generate a talking head; none of them films the actual product being used. The combination produces footage of the real thing — the one thing a scripted avatar can never fake.

MVP

A CLI that takes a URL plus scripted steps, drives a headless browser through them, retries failed scenes, and renders a 1080p Remotion timeline. Deliberately skip: AI narration voiceover, multi-language versions, and avatar presenters — ship silent-captioned footage first.

How it makes money

The moat

Synthesia, HeyGen, and Colossyan are text-to-avatar architectures: their models, pipelines, and enterprise contracts are built for presenter videos, not for driving a live product UI. Filming real product footage requires a different stack (agent harness + deterministic capture + scene retry), which is a new product, not a feature they can ship in a quarter — and it serves a different buyer (launching founders) than their L&D and marketing-avatar contracts.

Distribution

B2B wedge through launch agencies and Product Hunt / Hacker News launch concierges, plus self-serve for indie hackers and YC-batch startups; land on launch week, expand to every release.

Why it wins

Synthesia, HeyGen, and Colossyan generate avatar presenters reading a script — the output is a person talking, not your product working. TakeOne films the actual product being driven through real flows, so the demo shows what the buyer will really experience.

Risks

Flaky product UIs break takes mid-shoot. De-risked by filming against scripted sandbox environments with deterministic seed data and re-tryable scenes, so a broken step re-shoots instead of failing the whole video.

Build it with

• Alibaba OpenGCUA — full stack for generalist computer-use agentsOpen full stack for generalist computer-use agents that can drive a real product UI
• Holo4: H Company's open-weight models for computer-use agentsOpen computer-use model weights for the driving agent
• Kling 4.0 — global launch with professional-grade outputProfessional-grade video generation for transitions, polish, and narration visuals
• Jev 1.14.0 — TypeSafe's judgment layer updateTyped judgments as the decision layer: pick the best take across filmed attempts

Repo to start from

takeone — CLI plus agent harness that films a product demo by driving the real UI, then renders a launch-ready video.

Evidence

• Zelios: demo video pricing ($1,600, ~20 days)
• ContentBeta: how much does a product demo video cost
• The expensive mistake most founders make with product demos
• openvidstudio: MCP-driven demo videos (related prior art)