hypit-ai/hypit

What it solves

Hypit allows users to rapidly clone and iterate on viral video formats. Instead of manually editing a video from scratch, it provides a system where AI agents can treat video production as a programmable workflow. This enables the creation of hundreds of variants of a single successful video structure—such as changing the narrator, the product, or the language—without needing to rebuild the entire composition.

How it works

Hypit provides a language and system for AI agents (like Claude Code or Codex) to create videos. It transforms a reference video into a complete, editable workflow consisting of footage, captions, B-roll, and effects. These elements are anchored to words rather than fixed timestamps, allowing the timing to re-flow automatically when content changes.

Users can start by providing a reference video for the agent to clone, using a template, or describing a video from scratch. The system can integrate with various generation models for assets, but it can also render visuals using front-end code (code-rendered visuals) without calling external APIs.

Who it’s for

• Content Creators: Those making TikToks, Reels, or Shorts who want to scale viral formats.
• Performance Marketers: Users creating paid social ads who need to ship dozens of hook variants for A/B testing.
• Affiliate Marketers: People managing TikTok Shop videos where the same high-converting format is used for different products (SKUs).
• Developers: Those building AI-driven video automation tools.

Highlights

• Word-anchored timing: Captions and effects are tied to words, not seconds, ensuring seamless adjustments when scripts change.
• Agent-native workflow: Designed specifically for coding agents to manage the environment, generate materials, and build compositions.
• Pluggable components: Components like hosts or captions can be swapped independently without affecting the rest of the video.
• Code-rendered visuals: Ability to render visuals via code locally, reducing reliance on expensive generation models.
• Open source: No per-render fees or watermarks.