claude-video-effects
51 video effects for Claude Code: drop in a talking-head clip, type /effects orbit. Face tracking, behind-your-head layers. By @vibha.in.progress
翻訳待ち
A free Claude Code plugin that puts a full studio process behind AI-made brand videos. A 15-minute brand interview, then every video ships in your look.
翻訳待ち
プラグインGitHub @derrtaderrオンライン所有者未確認
Have you tried it? One tap is enough. No sign-in needed.
リポジトリ情報はまだ収集していません。
No updates detected yet. GitHub releases will appear here.
A free Claude Code plugin that stops AI-made brand videos from all looking the same. Write your brand down once, and every video goes through a six-step director's process before it renders.
翻訳待ち
Loading the full README.
Read the full README ↗Two videos for one fictional brand, Fern and Field, each directed end to end by the plugin from the same brand file. The second had to look different from the first because the brand's style ledger said so. The GIFs are silent previews; the MP4s with sound, and every stage behind them, live in examples/fern-and-field.
Install it with two commands inside Claude Code. In Claude Cowork, add it from the plugins screen instead.
/plugin marketplace add derrtaderr/brand-video-director
/plugin install video-direction@video-direction
You can one-shot-prompt a brand video in a minute, and it will look like every other one-shot-prompted video, because nothing forced a look decision. This plugin is the director's seat you skipped. It runs a staged studio process (treatment, style, boards, styleframes, build, QC) with a gate at each stage, so the look, timing, and sound are decisions on the record instead of the model's statistical average. It directs; HyperFrames renders.
Motion is demanded, not hoped for. Every beat names its camera move, entrance, state change and transition from a motion vocabulary; a motion critic measures the composition before it renders and fails it under the archetype's floor; and each new video has to reach for moves the brand has not used in its last two. The result is videos that read as designed rather than assembled.
Each frame below is a decision the process forced before the final render.
The style ledger then pushed the second video away from the first on every axis they
share: captions instead of a voice, panels that hold while the content moves instead of a
traveling camera, and the accent tracing a path instead of marking a word. Read the ledger
at examples/fern-and-field/style-ledger.md.
Run the two commands at the top of this page inside Claude Code, or add it from the plugins screen in Claude Cowork.
Working from a local clone instead? Point the first command at the folder:
/plugin marketplace add /path/to/video-direction-plugin
The install copies everything into your own Claude config, so it keeps working even if you delete the downloaded folder. It runs entirely on your machine and your Claude subscription.
/brand-init. A short taste-by-comparison interview — mostly A/B picks (world,
type, accent, audio, delivery), about 15 minutes — writes your motion-brand.md and
renders one proof styleframe in your new brand so you see "that's MY brand" before you
make a real video. (Prefer to do it by hand? Copy templates/motion-brand-template.md
to motion-brand.md and fill it in.) It is the source of truth for every video.video-direction skill runs treatment →
style → boards → styleframes and presents a styleframe A/B pair.Videos render silent by default, with a one-line note on the render telling you how to score it. The plugin never blocks a render to go source a track, and never ships silence by accident without saying so. Two ways to add sound:
video-work/<slug>/composition/audio/ and it scores the render. No key, no account.npx hyperframes tts, Kokoro) and background music (MusicGen). npx hyperframes doctor lists both; they light up once the local dependencies are installed. Free, and
nothing leaves your machine.ELEVENLABS_API_KEY in your
environment and the skill generates studio voiceover or scored music for you, masters it
to broadcast loudness, and drops it into the piece before the render. It runs locally in
your own session against your own key. The key is never bundled with this plugin, never
read from a file, and never sent anywhere but ElevenLabs. No key set means this tier
stays off and the free local path is the default. Any other provider works the manual
way too, generate the track yourself and save it into composition/audio/.The house rule — no video ships silent by accident — is taught, not enforced: your
motion-brand.md Audio section wins. If it permits silent, a silent render is correct.
Next: set your audio default in /brand-init (the Audio question), then ask for a video.
Once your motion-brand.md exists you don't re-run setup to make more videos — just ask
for one in plain language. The video-direction skill detects the brand file, skips
straight to the treatment, and reads your style-ledger.md (/brand-init writes it
beside the brand file; the skill appends a row per shipped video) so video 2 varies the
variant axes and never repeats video 1's look.
To change the brand itself, re-run /brand-init. With a brand file already present it
never silently overwrites: it offers a targeted edit (re-ask just one section —
constants, roster, selection, audio, or delivery) or a full regenerate, and regenerate
only proceeds after you type the word regenerate to confirm.
Next: ask for your next video, or run /brand-init and pick edit to adjust one section.
npx hyperframes ….Preflight: run npx hyperframes doctor once before your first render. It checks
Chrome, ffmpeg, and Node and names the exact fix for anything missing. The first npx
call may pause ~30s while it fetches — that is normal, not a hang. If that very first
fetch errors instead (e.g. a transient npm ETARGET), just re-run the command — a
cold-cache flake, not a broken install.
With that green, run /brand-init to build your brand and make your first video.
From the plugin root, node --test 'skills/video-direction/scripts/**/*.test.mjs' 'tests/*.test.mjs' runs the skill's scripts and the README contract.
Early, but the core loop works. The video-direction skill (the studio process) and
/brand-init (the taste-by-comparison brand interview, ending in a proof styleframe) both
run today. Next is a film that runs three different brands through the same process, side by side.
To start now, run /brand-init.