HavenSkills

Claude Video Skills

Build generated videos, organize media and render a finished composition.

Skills
6 skills

How to choose

Pick an authoring framework before choosing a renderer. Set the duration, output size and source assets up front.

Filter

All skills in this category

hyperframes

Create timeline-based video compositions from HTML and web assets.

by heygen-com
npx skills add heygen-com/hyperframes --skill hyperframes
Author
heygen-com
Repo
GitHub

hyperframes-cli

Initialize, preview and render HyperFrames projects from a terminal.

by heygen-com
npx skills add heygen-com/hyperframes --skill hyperframes-cli
Author
heygen-com
Repo
GitHub

media-use

Prepare media assets for a HyperFrames composition and keep track of their use.

by heygen-com
npx skills add heygen-com/hyperframes --skill media-use
Author
heygen-com
Repo
GitHub

hyperframes-registry

Find and integrate reusable HyperFrames blocks and components.

by heygen-com
npx skills add heygen-com/hyperframes --skill hyperframes-registry
Author
heygen-com
Repo
GitHub

remotion-best-practices

Build React-based video compositions with frame-aware animation and media.

by remotion-dev
npx skills add remotion-dev/skills --skill remotion-best-practices
Author
remotion-dev
Repo
GitHub

remotion-captions

Parse, time and render captions inside a Remotion composition.

by remotion-dev
npx skills add remotion-dev/skills --skill remotion-captions
Author
remotion-dev
Repo
GitHub

What video skills are for

A video skill gives an agent a composition model instead of a rendering wish. The hard part of generated video is not making frames — it is timing: which element appears on which frame, where the caption sits relative to the speaker, and whether the export keeps the audio in sync. These skills supply that model, usually as HTML or React with a frame-accurate timeline.

There are two frameworks on this page, and the first decision is which one you are in.

HyperFrames or Remotion

HyperFrames (heygen-com/hyperframes) builds timeline compositions from HTML and web assets. Four skills, in the order you actually use them: hyperframes for the composition itself — timeline, layers, animation; hyperframes-cli to initialise, preview and render from a terminal; media-use to prepare assets (video, audio, images) and keep track of which asset each layer uses; hyperframes-registry to pull in reusable blocks instead of rebuilding a lower third for the fourth time. If you already write HTML and CSS, this is the shorter path.

Remotion (remotion-dev/skills) builds React compositions. remotion-best-practices covers frame-aware animation and media: useCurrentFrame, interpolate, sequencing, and loading assets with <Img>/<Video> instead of plain tags. remotion-captions parses, times and renders captions inside the composition. If your team already writes React, this is the one — the animation model is the same one you use in a browser, just driven by frame number.

Do not mix them mid-project. Pick one, because a composition written for one will not render in the other.

Where these skills break in practice

  • Duration decided last. A composition is built for 30 seconds and the script runs 46. Fix duration, output size and source assets before the first frame is written.
  • Fonts that do not exist at render time. The preview uses your system fonts, the render container does not, and the export falls back to Times. Load fonts inside the composition or bundle them.
  • Captions timed to the script, not the audio. The caption file is generated from text, so every pause drifts it further out of sync. Time captions against the rendered audio.
  • Heavy assets loaded per frame. A 4K clip referenced on every frame makes the render take hours. Load once, reuse the reference.
  • Absolute timing copy-pasted. Changing one clip's length by two seconds shifts everything after it. Drive timing from a single source — a data file or a duration variable.

The check that catches most of it

  1. Render a short sample — 5 to 10 seconds — before the full render. Timing problems appear in the sample, and the sample takes seconds.
  2. Watch it with sound. Subtitle position, audio level and clipping are invisible in a frame grab.
  3. Check the last frame. Off-by-one frame errors show up as a truncated final caption or a fade that never completes.
  4. Compare the rendered duration with the intended duration. A file that is three frames short is usually a rounding default, not a bug in your content.

Which page to read next

Design and UI covers the visual layer these compositions borrow from — type scale, palette, spacing — and its advice applies to video frames unchanged. Coding skills covers the React toolchain Remotion needs.

How to evaluate the result

Render a short sample and check timing, subtitles, audio levels and visual clipping.

How to check the result

Watch the first three seconds and the last three seconds at full resolution. The first frame should show the intended scene, not a blank canvas; the font should be the one named in the composition, not a fallback; captions should still land within 0.2 seconds of the spoken words at the end of the cut.

Frequently asked questions

How do I install one of these video skills?

Use the command beside the skill, select a supported agent, then reload the agent and confirm it can see the skill in your project.

Do I need HyperFrames and Remotion both?

Choose one authoring framework per project. HyperFrames supports HTML and CSS compositions and a local render CLI. Remotion supports React compositions. Add caption or media skills when the workflow calls for them.

Why does the rendered video show missing images or the wrong font?

The renderer runs a headless browser, so anything loaded from a remote URL or a font that is not installed locally can fail silently. Copy assets into the project, reference them relatively, and bundle or install every font before rendering.

Can these skills add captions to a video that is already rendered?

remotion-captions works inside a Remotion composition, so the video needs to be rendered from that composition rather than post-processed. Burn-in after the fact needs a separate editor.

Which skill should I install first for a talking-head video with subtitles?

Start with the authoring skill for your format (hyperframes or remotion-best-practices), add remotion-captions if the subtitles are rendered into the frames, and add media-use or hyperframes-cli only when the project reuses assets or needs a local preview.

Do I need a GPU?

Not for these frameworks at typical resolutions — they render in a headless browser. Very long renders are a CPU and time question, not a GPU one.

Can it generate the video content itself?

It composes, animates and renders. Footage you supply with media-use, or generate elsewhere and bring in as an asset.

Why do my captions drift?

They were probably timed from the script text rather than the audio. Use remotion-captions against the rendered audio track.

What should I set before the first prompt?

Duration, output size, and the list of source assets. Those three decide everything the agent writes next.