Your recording becomes a video edited by an agent

✂️ Transcript-driven cuts

Removes silences, stutters, false starts and retakes with word-level timing. Every proposed cut is reviewed in context before it is applied.

🧩 Motion graphics at the right moment

Cards with CSS tokens and text slots, anchored to the exact phrase being spoken. A validator checks that each element comes in the moment you talk about it.

🎞️ From lesson to showreel

Full editing of lessons, reels and 9:16 Shorts, plus 15-second brand showreels cut to the timing of the music.

Transcribe

Word-level timing: it is what makes each graphic come in at the right moment and lets the agent understand the story. ElevenLabs Scribe or local Whisper.

Cut

Dead air and mistakes go before any design, with cut-silences and then cut-mistakes. Useful even if you will finish in your own editor.

Plan the beats

"When I say X, show Y." The clearer the agreement, the closer you get to what you want. video-storytelling and hyperframes-video-beats.

Use and create skills

Anything you repeat becomes a skill. After each round, say what you liked and what you didn't, and ask for the skill to be updated.

Verify in a loop

The agent renders, extracts frames, looks, fixes and only delivers after a few passes. You get version 5, not version 1.

Reference → skill

Liked a video? Measure it with analyze-reference.mjs, ask for an analysis of why it works and turn it into a skill. That is how motion-showreel was born.

🎯 Anchored to speech

Each element tied to a phrase, with a position and a glass card for readability.

When I mention [A] and [B], show [A] where I
point on the left and [B] on the right, with
animated logos and a glass card behind. When I
say "[thesis]", show the words below,
highlighted. Check the render before delivering.

📸 Prove it live

The agent fetches the material: records the screen, takes screenshots, uses real images and generates whatever is missing.

When I talk about B-roll, screenshots or image
generation, do it for real: record the screen,
take the screenshot, generate the image, and show
it on the sides with a background that keeps it readable.

🖍️ One recording, three styles

Hand-drawn whiteboard; course style with a circular camera crop and large bullet points; or a dynamic short with B-roll.

Edit [recording] as a lesson. The camera starts
full screen and then sits in a circular crop.
Minimalist background, a few large bullet points
and simple AI-generated images.

🎉 Sizzle from a folder

A folder of event recordings becomes a reel with the quotes that show the audience's energy.

Look at everything in [folder], transcribe the talks
and build a [duration] sizzle telling the story of
the event. Show me the quote selection first.

🛍️ Product video

Real catalog images, photos generated from them and short videos of those photos, timed to the music.

Make a [duration] video for [product] with
real catalog images. Where motion is missing,
generate a photo and video (Kling via my subscription).
Confirm before spending credits.

✨ Brand showreel

The motion-showreel skill: a motif that transforms into chapters, a HUD, cuts on the timing grid and a logo lockup.

Use motion-showreel for a 15 s showreel
of [brand]. Measure the timing grid, map the
chapters to brand objects and show me the
storyboard before building.

Local tools

Make sure they are on your terminal's PATH.

# versions
node -v   # 22 or later
ffmpeg -version
ffprobe -version

Assistant

Codex (desktop app or CLI) or Claude Code. The skills are discovered automatically when you open the folder. For full edits, use high effort.

# Claude Code, inside the folder
claude
# Codex: New project → Local → kit folder

Services (optional)

Transcription: ElevenLabs Scribe or local Whisper. Generated video: your Kling subscription via the kling CLI; Kie.ai only as a paid alternative.

# .env at the root
ELEVENLABS_API_KEY=your-key
# Kling: no key, OAuth login
kling login