Price terms as published

Each line is the vendor’s own wording, split from the price capture checked on 2026-10-24. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.

2 price terms for huashu-motion
#Plan / termChecked
1Free (MIT open source)2026-10-24
2you pay only your coding agent's own token/API costs2026-10-24

One-liner: huashu-motion is an open-source Agent Skill that turns your coding agent into a video director: say “make a 30-second launch film for my product, here is the site”, and the agent reads the product, offers three creative directions, writes the film, composes an original score, then opens a browser editor where you click on any frame and tell it what to change — it edits only that element and refreshes the preview in place.

Everything below was checked against the official repository (GitHub API + raw README, fetched 2026-10-24), the author’s launch write-up on NetEase, an AlphaSignal news story, and the author’s X thread.

Verified facts (GitHub API + raw README + third-party coverage, 2026-10-24)

Attribute Value Source
GitHub stars 1,546 (150 forks) alchaincyf/huashu-motion via GitHub API, 2026-10-24
License MIT GitHub API license field + LICENSE file
Created / last push 2026-10-18 / 2026-10-22 GitHub API
Language / stack JavaScript; renders via your installed Chrome/Chromium/Edge, exports with FFmpeg GitHub API + README “需要什么”
Install npx skills add alchaincyf/huashu-motion, or tell your agent “install huashu-motion” README “安装”
Compatible agents Claude Code, Codex, WorkBuddy, Grok Bot, Muse, Dots — anything that speaks the Agent Skills protocol README badges + install section
Core workflow Read product → propose 3 directions → shot list → compose score → browser editor with point-to-edit README “能做什么”
Point-to-edit Press C, click a frame, write a comment; the agent zooms into that frame, edits only that element, replies in plain language, preview refreshes in place; ⌘Z undoes any version README “指着画面改”
Scoring Agent writes a score in ABC notation (plain-text music); played by sampled grand piano, strings and drums — real instrument samples require om samples build (about 55 MB, CC0/CC-BY), otherwise synth tones; all composed locally, no paid APIs README “先写谱，再拍片”
Voiceover (optional) macOS system voices free; Kokoro for Chinese narration (needs uv + Python 3.12, ~325 MB model); word-level sync needs whisper-cpp (brew install whisper-cpp) README “需要什么”
Outputs MP4, MOV (transparent), WebM, GIF, or a shareable page with player + source; om share bundles a ≤10 MB GIF + caption + README header code README “能做什么”
Requirements Node 22+, Chrome (or Chromium/Edge), FFmpeg; tested on macOS only; Codex sandbox may need manual approval for local preview/browser README “需要什么”
Cost example Author’s 22-second DeepSeek V4.1 Flash film with music and SFX: about ¥1.95 (cache hit 99.4%) README “不依赖Opus” + AlphaSignal headline “$0.27”
Launch coverage Author’s NetEase article (Oct 2026), AlphaSignal news story, X thread Third-party, fetched 2026-10-24

This is an official-terms entry: we have not run huashu-motion ourselves, so there is no score. The one concrete cost figure above is the author’s own published example, not a guarantee.

Getting it running

Install is one command: npx skills add alchaincyf/huashu-motion. The skills installer (from Vercel Labs) detects which agent you are using and drops the skill into the right directory; add -a claude-code or -a codex to force a target. You can also clone the repo and symlink huashu-motion/skills/huashu-motion into ~/.claude/skills/ (or ~/.codex/skills/) — the skill’s bin/om calls the CLI tool inside the repo, so editing the repo edits the skill and there is only one copy on your machine.

Before the first film, run om doctor — it checks Node, browser, FFmpeg, and sample availability. Optional extras: om samples build downloads ~55 MB of CC0/CC-BY licensed piano/string/drum samples; for Chinese voiceover, Kokoro prints a one-time install command (uv + Python 3.12, ~325 MB model); for word-level lip sync, brew install whisper-cpp. Then the intended usage is plain language: “use huashu-motion to make a 30-second launch film for my product, the site is xxx.com”. The skill also handles data storytelling (a CSV becomes a 30-second explainer), voiceover films from your own recording, adding packaging overlays, vertical 1080×1920 versions, and GIF/transparent exports. om share then produces a ≤10 MB GIF plus a share.md with a caption and README header code.

What it costs

The skill itself is free (MIT), and scoring/composition happens entirely locally — the README states the score is performed by local sample rendering with no paid API calls. What you pay for is the coding agent’s own usage: your Claude Code/Codex subscription, or pay-per-token providers. The author’s published example is a 22-second scored film made with DeepSeek V4.1 Flash (Oct 2026 model) that cost about ¥1.95 (≈$0.27 per AlphaSignal’s headline) with a 99.4% cache hit — that number is one run, cache-warm, and will vary with your prompt history and model. Optional licensed samples and voice models are free to download; the only hard requirements are free tools (Node, a browser you already have, FFmpeg).

How it differs from alternatives

The nearest neighbors on this site are Remotion (a React framework for programmatic video, ~60K stars) and video-use (an MIT skill that edits existing footage with an agent). huashu-motion sits differently on both axes:

  • Remotion is a developer framework: you write React components that render frame-by-frame, and you own the whole pipeline. huashu-motion is the opposite — you give a sentence, the agent writes the shots, the copy, and even the score; the browser editor is your control surface instead of code. Choose Remotion if you want deterministic, versionable video code; choose huashu-motion if you want a launch film from a product URL without writing video code.
  • video-use edits existing talking-head/montage footage (transcribe → EDL → render). huashu-motion creates a film from nothing: motion graphics, kinetic typography, data storytelling, original scoring. Complementary rather than competing — huashu-motion for the launch cut, video-use for cleaning up real footage.
  • Against paid text-to-video generators (Kling, Runway, Sora-class), huashu-motion is deterministic-code output with an editable timeline, not a diffusion render — different look, and cost is agent tokens rather than video credits.

Where to stay honest

  • “One-click” means one sentence to an agent you already run. If you do not have Claude Code/Codex, the friction before the first film is meaningful.
  • macOS-tested only per the README; Windows/Linux may work but are not verified by the author.
  • Quality scales with direction. The author shows one-sentence films reaching 90+ via a second reviewing agent, but also shows the most polished film took an hour and a half and ten versions.
  • The ¥1.95 cost is the author’s warm-cache example, not a price list; your bill depends on model choice and cache state.

Who it’s for

  • Indie makers who already run a coding agent and want a launch film this week instead of hiring motion design
  • Agent tinkerers who want the “structured output beats raw generation” pattern applied to video + music
  • Teams making data explainers or product teasers who want editable, deterministic output they can tweak frame-by-frame
  • Not for: non-developers with no agent setup, Windows-first workflows (until tested), or teams needing photorealistic actor footage — this is motion graphics, not live-action generation

Sources