claude liam brutalist skill explainer
Explains how the explainer skill's two-axis shot system and slot contract let mixed media, archive photos, Manim fragments, and footage, read as one film under a single treatment.
Every other explainer skill in this video toolkit inherits from one parent, simply called explainer, not the AI-branded cut, not the CLI cut, not the deep cut. Reading its SKILL.md directly answers what it actually is and how it decides what ends up on screen.
The easy mistake
The first assumption most people make is that this skill defines a look: cream paper, editorial serif type, hand-drawn rings. It doesn't. It's a pipeline. It launders any source thrown at it, an archive photo, a Manim fragment, a public domain reel, through one consistent treatment. The sources never match each other. The treatment does. That's why a film built from wildly different source material still reads as one coherent film.
The anatomy
A skill in this toolkit is a folder Claude reads before working, and the explainer folder is larger than most because it governs everything downstream. SKILL.md is the core doctrine. A motion reference file holds a pantry of seven motion languages. A Remotion reference file specifies the annotation plane. An equations reference file provides the equation-tangent template. A scripts folder holds the runtime: the audio maker, the pantry cropper, the compiler, the outro machine. Every other explainer skill in the toolkit extends this one.
The pipeline
The flow runs: script becomes beats, beats become a Kokoro voiceover, free and local, which serves as the master clock. Media flows in per beat from four sources: Manim fragments, treated stills over a Ken Burns pan, public domain footage, and motion annotations. A compile step conforms every slot to the audio window and produces one watchable master. That's the entire loop; everything else is a modifier layered on top of it.
The laundering thesis
The skill file states its own thesis directly: it is not a style, it is a laundering function. Any source fed into it, an archival photo, a generated still, a public-domain reel, a Manim fragment, a motion card, passes through the same treatment: desaturated by roughly 80 percent, contrast raised to about 1.15, seated on a scanned newsprint texture. That single shared treatment is why mixed media reads as one film rather than a patchwork of different visual sources.
Two independent tags per beat
Every beat carries two separate tags. Shot type is the presentation form, still, footage, document, graphic, composite, or title card, and it gets locked at the planning stage and never changes afterward. Shot source is the provenance, whether it came from an archive, an AI generator, or a user's own file, and it's late-bound and swappable. Keeping these two tags separate is deliberate: collapse them into one, and every media swap becomes a full re-edit. Keep them separate, and swaps are free, drop a new file into the media slot and only that one slot recompiles.
The slot contract
Every beat on the timeline resolves to one conformed MP4 per beat, and the precedence at compile time runs: a hand-added MP4 in the media folder beats a Manim-rendered beat, which beats a still image with Ken Burns motion, which beats a placeholder slate. That ordering means the very first compile pass is always a watchable film, with slates standing in for anything not yet supplied. A slate is a normal production stand-in, not a defect; it marks a slot to be filled later. Changing a slot's input recompiles only that one clip, leaving the rest of the reel identical.
When a still earns its place
The skill's sharpest rule concerns when a still image belongs in the film at all: a still exists for exactly one reason, because it is the evidence, the actual document or record the argument turns on, not because it looks appropriate. The test is a single question: could this still be swapped for any other still of the same general subject without changing what the beat proves? If yes, it's decoration and should be cut, letting a Manim or Remotion beat carry the idea instead. A pantry still is the most expensive kind of beat in the system, worth paying for when it's genuine evidence, never worth it as atmosphere.
Key takeaways
- The explainer skill is a compositing pipeline that applies one shared visual treatment to any source, not a fixed aesthetic style.
- Kokoro-generated voiceover is the master clock; media is conformed to fit the audio, not the other way around.
- Shot type (presentation form) and shot source (provenance) are tracked separately so media can be swapped without a full re-edit.
- The slot contract's compile precedence guarantees the first pass is always a watchable film, using slates as placeholders.
- A still only belongs on screen when it is irreplaceable evidence for the beat's claim; otherwise it should be cut.
Who this is for
Anyone building explainer videos in this toolkit who wants to understand the parent pipeline every specific explainer variant inherits from, and viewers curious how mixed-media documentary footage can read as one consistent film.
Full transcript(auto-generated, with timestamps)
[0:00]Aloha, this is Liam in forbear. Today we tear down the skill every other explainer inherits from the one just called explainer. Not the AI cut, not the CLI cut, not the deep cut, the parent. What is it really? And how does it decide what ends up on screen? Here is the mental model. The easy mistake, the one every fellow makes first is thinking Vox is a look. Cream paper editorial serif handdrawn ring. It is not. Vox is a pipeline. It laundered any source you throw at it. An archive photo. A man fragment. A public domain reel through one treatment. Sources never match. The treatment does. That is
[0:32]Why the film reads as one film. First, a skill is a folder Claude reads before working. The explainer folder is bigger than most because it governs everything downstream. Skill MD is the doctrine. Motion MD holds the pantry of seven motion languages. Remotion. MD is the annotation plane spec. Equations. MD is the equation tangent template. The scripts folder is the runtime belt, the audio maker, the pantry cropper, the compiler, the outro machine. Every other explainer extends this file. Here is the pipeline. Script becomes beats. Beats become a kakoro voice over free local and the master clock. The media flows in per beat from four sources. Manm
[1:06]Fragments, treated stills over ken burns, public domain footage, and motion annotations. Compile conforms every slot to the audio window and hands out one watchable master. That is the whole loop. Anything else is a modifier on top of it. First design decision and it is the thesis. Quote from the skill file vox is not a style. It is a laundering function. Any source you feed it an archival photo a flux still a public domain reel a manum fragment or motion card passes through one treatment. Desaturate about 80% contrast raised to about 1.15 seated on a real newsprint scan. That is why mixed media reads as a
[1:39]Single film. The sources never match the treatment does. Second decision. Every beat has two independent tags. Shot dot type is the presentation form. Is it a still a piece of footage? a document, a graphic, a composite or a title card. Type gets locked at the plan gate and never changes. Shot dots source is the provenence. Did it come from an archive, an AI generator or your own file? Source is late bound and swappable. Collapse the two and every media swap becomes a re-edit. Separate them and swaps are free. Drop a new file in media. The reel rebuilds only that slot recompiles. Third decision, the slot contract. Every
[2:13]Beat on the timeline is one conformed per beat MP4. The precedence at compile is a hand added MP4 in media beats a manm beats a still image with Ken Burns beats a slate. That means the very first pass is always a watchable film. A slate is a standard production standin, not a bug. It says fill this slot later. Change a slot's input and only that clip recompiles. The rest of the reel is bite forbite identical. That is why swaps cost nothing. Here is where the skill bites box law. A still exists for exactly one reason. The still is the evidence, the actual record, the document, the argument turns on.
[2:48]Princeton's actual admissions record, yes. A stock photo of a university? No. The test is one question. Could this still be swapped for any other still of the same subject without changing what the beat proves? If yes, it is decoration. Cut it and let a remotion or man beat carry the idea. A pantry still is the single most expensive beat in the system. It is worth paying for evidence, never for atmosphere. Verdict. The explainer skill is the chassis every other explainer bolts onto. It is a compositing pipeline, not a style. The Kakoro voiceover is the master clock and free. The twoaxis shot system separates presentation from prudent, so swap stay
[3:22]Cheap. The slot contract makes the first pass a watchable film with slates, and VOXW keeps every still that reaches the timeline loadbearing. Do this well and mixed media reads as one film. Break the last rule and you shipped a slideshow. Your turn. Paste this into clawed code inside a fresh reels folder. Run slate cut on the concept. Write a short script. Plan the beats, generate cakoro audio, and produce a watchable review cut with slates in every human media slot. Then read the film back and check three things. Does every bodybe have a specific job, a comparison, a mechanism, a claim on screen that only that beat
[3:55]Can do? Is every plan still a piece of evidence, not atmosphere? And if you swapped anyone still for another of the same subject, would the beat still prove what it claims? If it would, cut that still and let a remotion or man and beat carry the idea. That was the explainer skill. Lay them in for bear.
