video-shotcraft: an agent skill that storyboards product promos in Remotion
AI video skill for Claude Code & Codex — cinematic product videos with Remotion: 152 shot recipe cards, 209 motion previews, a production-ready template
At a glance
- What is it?
- video-shotcraft is a Claude Code and Codex skill built around 152 shot recipe cards, 209 motion previews and a Remotion template called Ink Press. It is a good fit if you already edit video or write React; it is the wrong tool if you need footage you did not shoot or a render pipeline that runs without an agent in the loop.
- Who is it for?
- Adopt video-shotcraft if you already work in React and Remotion and want an agent to storyboard and assemble a promo from a fixed shot library; skip it if your video depends on live-action footage, actor performances, or a render pipeline you cannot supervise.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 7 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap video-shotcraft tries to close: motion design without a motion designer
Most product teams can write a launch post. Fewer can produce a 36-second promo with camera moves, beat-synced cuts and sound design, because that work normally needs a motion designer and a timeline editor. video-shotcraft targets that gap by packaging motion design as an agent skill: you point Claude Code or Codex at your product, and the skill storyboards, animates and sound-designs a promo video using Remotion. The README describes the output as including real page captures, 2.5D camera moves, beat-synced cuts and film-grade SFX. The intended user is a developer or technical marketer who is comfortable in a terminal and willing to let an agent drive the assembly, not someone looking for a hosted editor with a drag-and-drop canvas.
How the shot recipe cards and Remotion components fit together
The library is the core of the project. The README's headline counts 152 shot recipe cards and 209 motion previews, and the changelog entry for 2026-08 states the library grew from 104 cards to 152 after 48 new cards were distilled from 209 candidate motions through eight rounds of frame-by-frame review against reference footage. Each card ships with a native Remotion component at a path like demos/<category>/<name>/<Component>.tsx. Those components are described as deterministic and driven by a normalized progress value t, which is the mechanism that makes them reusable: a shot does not know its own duration, it only knows how far through it is, so the same component can be retimed without rewriting it. The README also states the cards are de-branded, with neutral placeholder copy and a single swappable ACCENT color variable. That is a deliberate constraint. You get a consistent visual system across shots, and in exchange every shot inherits the same accent-driven palette until you change it.
Installing it into Claude Code or Codex
The README gives three installation routes. The most direct is to tell your agent, in plain text: Install this skill for me: https://github.com/Vincentwei1021/video-shotcraft. The agent then clones the repo and links it into your skills directory. The second route uses the skills CLI: npx skills add Vincentwei1021/video-shotcraft. The third is manual: git clone the repository, cd into it, then symlink it, either ln -s "$(pwd)" ~/.claude/skills/video-shotcraft for Claude Code or ln -s "$(pwd)" ~/.codex/skills/video-shotcraft for Codex. Once installed, you make requests in natural language and can name specific cards, for example asking for the deck-deal-flyin and row-embed shot cards, or requesting a product close-up inspired by spotlight-hero-card. If you do not name a card, the README states the skill introduces the built-in video template first and asks whether to use it. The Motion Workbench is opened after delivery with node workbench/scripts/open.mjs <project>.
Ink Press: what the bundled template actually pins down
Ink Press is the shipped template, described in the README as validated and complete: 36.2 seconds, 1920x1080, 30fps, 10 shots in a paper-ink-amber style, with 2.5D real-page camera moves, title cards, transitions and a fully pinned cinematic SFX pass. Those numbers matter more than they look. A fixed frame rate and resolution means the shot components were authored against one canvas, and a fixed duration means the timing of cuts and SFX was tuned to a specific runtime. Using Ink Press as-is is the low-risk path. Reusing individual cards inside your own composition is a different job, and the README points to demos/README.md for the wiring snippet rather than claiming the cards drop in without setup.
The Motion Workbench and the frame-parity claim
The 2026-09 entry describes a CapCut-style workbench that opens after delivery. The film is decomposed into shot, transition, caption and SFX tracks exactly as authored. From there you can select a shot and edit copy, font sizes and colours in what the README calls a schema-driven inspector, move, trim or speed-ramp clips, drag any of the 216 demo motions in from the library, and export with Remotion. The README states preview and render are frame-identical and describes this as pixel-parity verified. That is a strong claim and it is the one I would test first, because a browser preview that drifts from the rendered output is the classic failure mode of any web-based video editor. The workbench guide is in Chinese, and the integration contract lives at references/workbench.md.
Where the approach breaks down
The dependency on Remotion is the load-bearing constraint, not a detail. Every shot is a React component rendered through Remotion, so your Node and Remotion versions have to satisfy the template's package.json before anything renders, and a Remotion upgrade can break the whole card library at once. The skill also assumes an agent is in the loop for authoring; the workbench helps after delivery, but it is not a replacement for the storyboard step. The visual output is bounded by the card library: 152 cards is a large set, and it is still a set, so a promo that needs a shot nobody wrote means writing a new Remotion component by hand. The README also notes that the JianYing export was verified on JianYing Pro 11.2 for macOS, which is one platform and one version, and the workbench guide is only in Chinese, so non-Chinese readers are working from the integration contract and the screenshots.
Remotion versus a timeline editor, and the JianYing escape hatch
The obvious alternative is a conventional editor such as CapCut, Premiere or DaVinci Resolve. The difference in approach is structural. A timeline editor stores decisions as edits on a track, which you can nudge frame by frame but cannot parameterise; Remotion stores them as code, which means a shot can be retimed by changing a duration prop and a colour can be changed in one variable across every card. That is why the project can offer a single ACCENT variable and a schema-driven inspector at all. The cost is that nothing exists until it renders, and a broken component fails the whole build rather than one clip. The project partly bridges the two worlds: the 2026-08 entry describes exporting the finished film as an editable JianYing draft, with the plate cut per shot for retiming, reordering and grading, captions rebuilt as native text tracks with editable content, size and colour, and SFX and BGM on separate audio tracks. That export is the pragmatic answer for teams that want code-driven assembly and then hand off to an editor.
Maintenance cost and what the Apache-2.0 licence does not cover
The repository is TypeScript, licensed Apache-2.0, not archived, and the last push recorded is 2026-09-09. The release history is thin: showcase-media on 2026-09-01 and gallery-media on 2026-07-26 are media asset drops, not library releases, so the 152-card library and the workbench arrived through commits rather than tagged versions. If you pin to a commit, budget for the fact that there is no version number to pin to. Licence-wise, Apache-2.0 covers the repository's own code and grants patent terms and a modification right, but the README says nothing about Remotion's own licensing, and Remotion is a separate project with its own terms that may apply to commercial use. Verify that separately. The same caution applies to any bundled media and to the fonts and SFX the template ships with, since the README does not enumerate their provenance.
Editorial conclusion
Adopt video-shotcraft if you already work in React and Remotion and want an agent to storyboard and assemble a promo from a fixed shot library; skip it if your video depends on live-action footage, actor performances, or a render pipeline you cannot supervise. Verify two things before committing: that your Node and Remotion versions satisfy the template's package.json, and that Apache-2.0 plus the Remotion licence terms cover your distribution, since the README states the project's own licence and nothing about Remotion's. The card library is the durable asset here, not the template.
Community notes