# Hypit: AI Agent Video Authoring with SVML Workflows > Hypit gives AI coding agents a language and system for producing video: drop in a reference video and the agent clones the full workflow in SVML, or start from a template or description. Generation models are optional; the same workflow can compile captions, motion graphics and code-rendered visuals without calling any AI image or video generation service. **hypit-ai/hypit** — Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views. - Repository: https://github.com/hypit-ai/hypit - Website: https://hypit.ai - Stars: 16,639 · Forks: 2,005 - Language: TypeScript - License: NOASSERTION - Published: 2026-09-13 · Updated: 2026-09-13 · Language: en - Canonical page: https://hysenlabs.com/projects/hypit-ai-hypit ## What Hypit Does and Who It Is For Hypit is a video authoring distribution for AI coding agents. Its primary use case is cloning the format of an existing video: drop in a reference video and the agent produces a complete workflow describing the footage, captions, B-roll, effects and transitions in SVML (Structured Video Markup Language). That workflow can then be varied to produce multiple outputs with different content but the same structure. Cloning is described in the README as the fastest entry point, not the only one. An agent can also start from Hypit's included templates or generate a workflow from a plain-language description of the video the user wants. The target audience is developers and content creators who want to produce multiple video variants programmatically rather than editing each one individually. Hypit supports Claude Code, Codex and other agent environments. The README states Hypit itself is free to use; charges come from the coding agent environment and any external model services used for generation. ## SVML: Workflows Anchored to Words Instead of Seconds SVML (Structured Video Markup Language) is Hypit's core abstraction. Instead of describing a video as a timeline with clips positioned at specific second marks, SVML anchors elements to words in a script. This means that when the script changes, the visuals, captions and effects attached to those words move with the text automatically rather than requiring manual repositioning. The README describes SVML as giving agents "a language and system to create video." The package.json exposes several internal packages that reflect this structure: narrative handles the script and word-timing layer, caption handles caption generation, composition handles the assembly of visual elements, browser-capture handles the final render using headless Chromium, and artifact handles output packaging. A video workflow in SVML is source code. It can be version-controlled, diffed and modified programmatically, which is why the agent can produce multiple variants from the same source: swap the narrator, change the rankings, replace characters. The examples in the repository demonstrate this with three clone variants per reference video. ## Installing Hypit and Running the First Workflow Install the Hypit skill globally with the skills CLI: ```bash npx skills add hypit-ai/hypit -g ``` This installs the skill into the configured agent environment. On first use, the agent checks whether the Hypit executable is present and helps prepare it if needed. The README states the video project can live anywhere on disk after the global install. Hypit uses HypiHub as its recommended hosted model service for generation tasks, but the README notes the user can substitute their own API or local models by telling the agent the service name and API documentation. Generation models are entirely optional: the README states explicitly that a workflow can compile captions, motion graphics and code-rendered visuals into a finished video without calling any generation model, avoiding those service charges entirely. The repository is at version 0.2.16, released on 2026-09-26. ## What the Examples Demonstrate The repository includes three working examples under examples/: a UGC tier-list format (ranking-football), a podcast clip format (podcast), and a street interview format (interview). Each example includes the original reference video, the SVML source file, three clone variants and production notes. The README describes the ranking-football example as a 20-second football tier list rendered using two Seedance 2 Mini 720p generation clips, ten B-roll panels from GPT Image 2, WhisperX word alignment, a sound-synced ranking board, color-box karaoke captions and animation rendered concurrently in 64 headless Chromium processes. The README gives the total cost as $1.15 for that example. The podcast example produced an 18-second clip using a split-screen interview layout, speaker-aware karaoke captions and a product-handoff moment, costing $1.07 according to the README. These cost figures come from the example production notes and reflect the generation model charges for those specific runs; they are not guaranteed for other projects. The examples/minimal-author-package/ and examples/semantic-composition/ directories provide simpler starting points for developers learning SVML authoring. ## Generation Models Are Optional A significant design choice in Hypit is that AI generation models are not required. The README states this directly: "Generation models are optional too: a workflow can compile captions, motion graphics and code-rendered visuals into a finished video without calling a generation model or incurring its service charges." This matters for two reasons. First, it separates the workflow authoring step from the cost of generation: a developer can iterate on the SVML structure and timing without spending money on image or video generation on every test run. Second, it means Hypit is usable for outputs that do not need AI-generated visuals at all, such as motion graphics, code-rendered animations and caption videos over existing footage. When generation is needed, Hypit works with HypiHub or any service the user configures by providing the service name and API documentation to the agent. ## Limitations and Cases Where Hypit Does Not Fit Hypit depends on an AI coding agent environment. It is not a standalone desktop application or a web UI. A user who wants to produce video without writing code or interacting with an agent cannot use Hypit directly. The SVML format is specific to Hypit. Workflows written in SVML are not portable to other video editing tools or timelines. If a project later needs to move to a conventional editor, the SVML source does not translate directly to that tool's native format. The license is listed as "SEE LICENSE IN LICENSE" in package.json, not as a standard identifier like MIT or Apache. The README states Hypit is free to use but does not summarize the specific redistribution or commercial use terms; those are in the LICENSE file in the repository. ## Maintenance Status and Package Structure The last push to the hypit-ai/hypit repository was on 2026-09-26. The repository is not archived and has recent releases: v0.2.16 on 2026-09-26, v0.2.15 on 2026-09-25, and v0.2.13 on 2026-09-23. The repository is a TypeScript monorepo under pnpm-workspace.yaml with packages split into separate directories under packages/. The package.json exposes thirteen sub-packages via the exports field: browser-capture, artifact, author-kit, caption, composition, endpoint-kit, generation, media, model-kit, narrative, program-space, runtime-kit and others. Skills for different agent environments live under skills/. A CONTRIBUTING.md file and AGENTS.md are present, indicating the repository is structured for external contributions. ## Conclusion Hypit is suited to developers and content creators who want to use AI coding agents to produce video at scale, especially when the goal is generating multiple variants of a single format. The prerequisite is familiarity with an agent environment such as Claude Code or Codex and a working understanding of how to configure model services. The LICENSE file governs redistribution and commercial use terms; the README states that Hypit itself is free to use but that coding agents and model services carry their own charges. ## FAQ ### What is SVML in Hypit? SVML (Structured Video Markup Language) is Hypit's markup language for describing video workflows. It anchors visual elements, captions and effects to words in a script rather than to specific time positions, so when the script changes, the attached elements move with it automatically. ### How do I install Hypit in Claude Code? Run npx skills add hypit-ai/hypit -g in a terminal. This installs the Hypit skill globally for the detected agent environment. On first use, the agent checks for the Hypit executable and helps prepare it if it is not already present. ### Can I use Hypit without a generation model? Yes. The README states that generation models are optional and that a workflow can compile captions, motion graphics and code-rendered visuals into a finished video without calling any generation model or incurring its service charges. ## Sources - [Official documentation](https://hypit.ai) - [Official README](https://github.com/hypit-ai/hypit#readme) - [Project repository](https://github.com/hypit-ai/hypit) - [Release notes](https://github.com/hypit-ai/hypit/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/hypit-ai-hypit