hypit-ai/hypit

Hypit – AI‑driven video‑cloning toolkit

What it is – Hypit is an open‑source framework that lets large‑language‑model agents (e.g., Claude Code, Codex) generate complete video productions. You give the system a reference video (or just a textual description) and it produces a workflow that describes every element of the final video – raw footage, captions, B‑roll, visual effects, and even code‑generated graphics. The workflow can be re‑run to create dozens or hundreds of cheap variants, making it useful for rapid creation of ads, TikTok‑style clips, podcasts, interviews, and localized versions.

How it works

• A SVML (Simple Video Markup Language) source file encodes the video structure.
• Hypit’s “skill” (installed via npx skills add hypit-ai/hypit -g) makes the workflow available to any compatible AI coding agent.
• The agent can:
  1. Analyse a reference video and emit an SVML description.
  2. Fill the description with assets (AI‑generated images, voice‑overs, captions, etc.) using optional models such as GPT‑Image 2, WhisperX, Google Video Intelligence, YOLOv8 AnimeFace, etc.
  3. Render the final video by driving up to 64 headless Chromium processes, all orchestrated from the SVML file.
• Because the workflow is declarative, you can edit or replace individual components (swap the host, change the language, replace B‑roll) without rebuilding the whole script.

Key features

• One‑click cloning – drop a video file, get a full production pipeline back.
• Mass‑variant generation – create 100+ inexpensive variants from the same workflow.
• Pluggable components – use built‑in modules, fork the library, or write custom ones.
• Zero‑cost rendering – a workflow can be rendered entirely with code‑generated assets, avoiding any paid AI API calls.
• Open‑source, no seat fees – the core engine is Apache‑2.0‑compatible; you only pay for any external services you choose to call.

Installation

# Install the Hypit skill globally (requires Node 22+, pnpm, TypeScript)
npx skills add hypit-ai/hypit -g

The first run will download the Hypit executable and set up the environment. Your video projects can live anywhere on your filesystem.

Typical usage (via an AI coding agent)

/hypit Clone this video: /path/to/viral.mp4

or

/hypit Make a ranking video that puts Hypit in S tier.

The agent will:

1. Generate or retrieve the SVML workflow.
2. Pull or synthesize assets (images, audio, captions).
3. Invoke the Hypit renderer to produce the final video.

Example outputs – The README showcases three demo families (UGC ranking, podcast clip, street interview). Each includes a reference video, three AI‑generated clones, and a cost breakdown (≈ $1 per video). The clones demonstrate swapping hosts, changing language, or altering product focus while preserving the original structure.

Use cases

• Paid‑social ads – clone a high‑performing ad, replace the product, and spin out dozens of variants quickly.
• TikTok/Shorts replication – turn any viral short into a reusable template for new content.
• E‑commerce & affiliate videos – generate daily product showcases from a single format.
• AI‑generated UGC/talking‑heads – automatic captions, word‑level alignment, and AI‑generated B‑roll.
• Podcast or interview highlights – split‑screen layouts, speaker‑aware captions, and animated overlays.
• Code‑rendered visuals – produce graphics entirely from front‑end code, eliminating external model calls.
• Localization – re‑run the same workflow with translated text to get multilingual versions.

Contributing – The project welcomes PRs for new components, documentation, examples, and translations. Contribution guidelines are in CONTRIBUTING.md; community discussion occurs on Discord, Telegram, and GitHub issues.

License – Distributed under the Hypit Open Source License (a permissive Apache‑2.0‑based license). Output videos belong to the user; any third‑party models or services used retain their own terms.

---

Hypit provides a bridge between LLM agents and full‑fledged video production, enabling rapid, low‑cost creation of viral‑style content.