Vincentwei1021/video-shotcraft

The problem

Making a short product promo usually means either handing a brief to a video editor and waiting, or wrestling with a timeline editor yourself. Generative video tools produce frames you cannot precisely control or edit afterwards. For developer-marketing work — a launch clip for a repo, a feature announcement, a README hero video — the source material is already on screen: your real UI, your copy, your brand colors. The gap is motion design skill, not assets. video-shotcraft addresses this by giving a coding agent a library of pre-tuned motion recipes plus a Remotion template, so the promo is drawn from your captured pages rather than prompted into existence.

What's different

The key distinction is that this is a skill for an agent, not an app you operate. You describe the film in natural language inside Claude Code or Codex; the agent captures pages, picks shot cards, storyboards, and renders. The shot library is the substance: 124 recipe cards across opening, camera, transitions, typography, UI entrance, interaction, data, light, rhythm and outro, each with a Remotion implementation and a gallery preview. Adjacent alternatives exist in different forms — Remotion itself is a coding framework with no recipe library; template-driven tools like Motion Canvas focus on explanatory animation rather than product capture; generative video models trade control for convenience. The trade-off here is that output quality is bounded by the recipe set and Remotion's code-drawn aesthetic, not by a diffusion model's imagination.

Two capabilities genuinely stand out: the named shot cards, which let you request a specific effect by name and get the same tuned result each time, and the export of the delivered film as an editable JianYing (CapCut CN) draft with per-shot clips and separate SFX/BGM tracks, which means the agent's output is a starting point, not a dead end. A browser-based Motion Workbench for post-delivery editing is also claimed.

Who it's for

Developers and DevRel/marketing engineers who already work inside Claude Code or Codex and need short, on-brand product videos without a motion designer. It fits teams comfortable reading and tweaking TypeScript/Remotion code. It does not fit people who want a zero-code editor, long-form narrative video, or footage-driven content — the output is UI-and-typography driven, and anyone expecting photorealistic AI generation will be disappointed.

Cost of adoption

The workflow assumes Claude Code or Codex; the README's install path is agent-driven or via npx skills add Vincentwei1021/video-shotcraft. Remotion is a Node/TypeScript dependency, so you need a JS toolchain and rendering compute. Apache-2.0 keeps commercial use open. Risks: the project is only about three months old, so the recipe catalog and export formats may still shift; the JianYing export is verified only on one platform and version; and the README is heavy on gallery imagery and sponsor content, with thinner detail on frame-QA behavior and failure modes.

Verdict

This is a brand-new project, not an established one: created 2026-07-19, roughly 97 days old, with 13.4k stars and a two-day trending run in late October gaining about 200 stars a day. Momentum is real but young, and the star count is far ahead of the visible maintenance history. The recipe-library-plus-agent-skill framing is a sensible answer to a real annoyance for developer marketers. Adopt it as a fast first-draft generator you intend to edit, keep the exported draft and the Remotion code as the durable artifact, and watch whether the card library stabilizes before building a pipeline on it.