Hypit

// Overview

Hypit gives AI coding agents — Claude Code, Codex and others — a language and a system for making video. You drop in a reference clip and the agent clones the whole workflow rather than just a script: footage, captions, B-roll and effects, all anchored to words instead of timestamps. The source format is SVML, a markup language that compiles to a finished video, so the output of a run is an editable, re-runnable composition rather than a single render.

Cloning is the fastest entry point but not the only one. You can start from the project's templates or simply describe the video you want and let the agent write the composition from scratch. Generation models are optional too: a composition can compile captions, motion graphics and code-rendered visuals into a finished video without calling a generation model at all, which is what keeps the code-rendered examples free of per-clip service charges.

The repository ships worked examples with their SVML sources and production notes — a ranking/tier-list short, a split-screen podcast clip, a street interview — each with several clones that swap the host, the language, the product or the rankings while keeping the structure that worked. The notes are concrete about the pipeline used: word-level alignment with WhisperX, speaker-aware karaoke captions, face bounding boxes for head-tracked captions, and dozens of headless Chromium processes rendering concurrently.

Hypit itself is free and open source with no seat pricing, per-render fee or added watermark; your coding agent and whatever model services you point it at are billed separately by those providers. HypiHub is the project's recommended hosted model service, but you can use your own API or local models by telling the agent the service name and its API documentation. It runs on Node.js 22.15+ with pnpm, is written in TypeScript, and is licensed Apache-2.0 with conditions.

// What it does

• SVML, a markup language for video that compiles a composition into a finished render
• Clone a reference video into a complete editable workflow — footage, captions, B-roll and effects
• Word-anchored timing, so rewriting a line re-flows the edit instead of breaking it
• Re-run one composition for many variants: swap host, hook, product, language or aspect ratio
• Pluggable components — swap the host without touching the captions; use the library, fork it or write your own
• Generation models are optional: captions, motion graphics and code-rendered visuals compile without any model call
• Runs as a `/hypit` skill inside a coding agent, which checks the environment and asks for the credentials a video needs

// Getting started

Hypit installs once as an agent skill. Video projects can live anywhere; on first use the agent checks for the Hypit executable and helps prepare it.

npx skills add hypit-ai/hypit -g

/hypit Clone this video: /path/to/video.mp4, and replace the ranking content with a comparison of Hypit (official website: hypit.ai) with other AI video products.

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

examples/ranking-football/reference.svml
examples/podcast/reference.svml
examples/interview/reference.svml

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it when one video format has to ship as dozens of variants — hooks, products, languages, aspect ratios
• Reach for it to turn a competitor's or your own winning short into a reusable template instead of a one-off edit
• Reach for it for podcast and interview clips that need split-screen layouts and speaker-aware captions wired automatically
• Reach for it when you want captions and motion graphics compiled from code, with no generation-model spend per clip

// How Hypit compares

Hypit alongside other open-source video generation tools AI/TLDR tracks, ranked by GitHub stars.