Hypit - AI Video Clone Production Setup¶

Source: hypit-ai/hypit via skills.sh - 19.0K combined installs across 1 indexed listing; new to this catalog in the Oct 24, 2026 evening sweep

GitHub: hypit-ai/hypit - 25,425 stars, GitHub reports the license as NOASSERTION, last pushed 2026-10-24, branch main, one skill at skills/hypit/SKILL.md in a 1,704 file repository

Category: Video Production / Content Automation

Quality Tier: 🟡 Beta - large and fast moving repository with clean auditor verdicts and a genuinely useful production skill, but the license is a custom one and the skill drives a separately installed executable plus paid model services

Hypit is a video production system for agents, and its claim is narrower than "AI video generation". Drop in a video and the agent clones the workflow rather than writing a script about it: the footage, the captions, the B-roll, the effects, the timing. The result is an editable composition you can re-run, not a one-off render, which is what makes the second half of the pitch work. One winning format becomes fifty hook variants the same day, or the same ad with a new product, price and call to action.

The architecture is the interesting part. Compositions are authored in SVML, a source format the repository renders live next to the source while you edit. Generated material and code-rendered visuals land in the same composition, so a video can mix model output with front-end code that renders locally without any generation API call. Components are pluggable, so the on-screen host can be swapped without touching the captions.

The economics matter too. Hypit itself carries no seat price, no per-render fee and no added watermark, and the outputs belong to you. What it does not provide is a model account: the skill installs authoring tools, and the generation services behind it are billed by whichever service you connect, with HypiHub as the recommended hosted option or your own API key for any other provider.

Installation¶

The README documents one install step:

npx skills add hypit-ai/hypit -g

The global flag is in the README because the skill is meant to work from any project directory, including an empty one. On first use the agent checks whether the Hypit executable is present and helps prepare it, then asks what video project it is working in.

There is no separate credential file to create by hand. The installation supplies authoring tools only, and the first session walks through the service choice: HypiHub as the integrated hosted service, or a provider you already pay for with its own API key. Service choice and credential setup are handled as two decisions rather than one.

Minimum environment stated in the README badges: Node.js 22.15 or newer and pnpm 10.33. Beyond that the agent does the environment check itself before it starts producing.

What It Provides¶

Skill	Installs	Use For
hypit	18,987	Directing and producing a video end to end: reading a reference, structuring the composition in SVML, choosing and connecting generation services, authoring components and rendering the finished piece.

No indexed listings fall below the threshold; the single listing appears above.

The skill is one large procedure rather than a menu of small ones, and its structure is worth knowing before the first run: establishing the project boundary, directing the material, preparing the work, adapting a reference, composing and refining, then standing responsibilities. It asks the agent to speak aloud as the work develops, to state what a setup step or a render is doing while it runs, and to bring the user into choices about private facts, spend and connected services rather than deciding quietly.

Why This Matters for Hermes Agents¶

Short form video is the one content format where "produce a variant" is a daily need and the manual cost is highest. An agent that can take a reference video, keep the structure that works, and regenerate only the parts that change turns a production task into a parameter sweep.

The design decision that makes this survivable for an agent is the editable composition. A single render is a dead end, because the next request is always the same video with a different product or language. Because Hypit writes a composition with named components and a source format you can read, the second request edits the composition instead of re-rolling the dice.

The second design decision is cost transparency. Code-rendered visuals are produced locally with no generation call, generated footage goes through whichever model service you connected, and the skill is instructed to surface the remaining preparation and cost before spending. For an agent working unattended, a spend boundary stated in the procedure is worth more than a cheaper model.

The caveat is maturity of terms rather than code. This is a young, fast moving project with a custom license, and the skill is a long procedure that assumes the model can hold a production plan across many turns. Run the first session supervised and read what it writes before you let it render on a schedule.

Usage¶

You say	What happens
"Clone this video and swap the ranking for my product"	The skill reads the reference, inspects its frames and speech, extracts the structure, and rebuilds it as an editable SVML composition with your product in place.
"Make a ranking video with us in S tier"	Starts from no reference, authors the composition structure and generates the material the piece needs.
"Give me 50 hook variants of this ad"	Reuses the existing composition and material, regenerates only the openings and keeps the body untouched.
"Turn this podcast into clips"	Builds split-screen layouts with speaker-aware captions and reaction overlays.
"Localise this video into ten languages"	Rewrites the lines and re-flows the timing in the composition rather than re-rendering blindly.
"Make a code-rendered intro, no generation calls"	Authors visuals as front-end components rendered locally, outside any model service charge.
"What will this cost before you render"	Reports the service choice, the material plan and the remaining preparation before spending anything.

Verification¶

The skill is a single path on main. Confirm it before installing:

curl -sL -o /dev/null -w "%{http_code}\n" \
  https://raw.githubusercontent.com/hypit-ai/hypit/main/skills/hypit/SKILL.md

Read the procedure before the first render, because it defines what the agent will ask you for:

https://raw.githubusercontent.com/hypit-ai/hypit/main/skills/hypit/SKILL.md

Two checks after install. The skill should appear in your agent's global skills directory, and its first run should stop to confirm the executable and the service choice before producing anything. If it starts generating material without telling you which service is connected, stop the session and read the environment references it ships with.

Security¶

skills.sh verdicts for sampled skills (verified October 24, 2026) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
hypit	Pass	Pass	Pass

All three auditors pass, and the repository ships no credential of its own. The risk sits where the auditors cannot see it: this skill connects to generation services you pay for and drives a locally installed executable. Its own text is explicit that credentials are connected per service, that HypiHub is one option among several, and that the skill supplies no generation account or model credits.

Treat two things as supervised. First, spend: a production run can call paid generation services many times, so set a budget with the agent before the first batch. Second, licence scope: third-party models and services carry their own terms, and the output rights the README grants cover Hypit itself.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified October 24, 2026: 18,987 combined installs across 1 indexed listing; 25,425 GitHub stars; GitHub reports the license as NOASSERTION; last pushed 2026-10-24. Counts drift over time.
The licence is a custom one. GitHub cannot classify it and the README links a Hypit Open Source License while the repository badge reads Apache-2.0 with conditions. Read the LICENSE file before commercial use.
The skill is a long production procedure, not a single command. It assumes a capable model and a session that can hold a brief across many turns.
Generation is billed by the model service you connect, not by Hypit. A batch of variants multiplies that cost.
Node.js 22.15 or newer and pnpm 10.33 are stated as the environment. Older runtimes fail at the executable step.
One author organization, a repository that changes daily, and one indexed listing. Pin the skill version you validated rather than tracking main.

Pro Tips¶

Start with a reference you can legally reuse and a product you control. The clone is faithful enough that the first test should not be a competitor's live ad.
Ask for the material plan before the render. The skill is written to explain which parts will be generated, which are reused and what the service calls will cost.
Keep the composition. The value of a run is the editable file, so commit the project directory and treat renders as outputs of it.
Use code-rendered components for anything with text or brand assets. They render locally, cost nothing, and survive a model service outage.
Run one variant end to end before asking for fifty. A structural mistake multiplies with the batch.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ