⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
explainer-board-brief
Decide WHAT goes on the explainer boards (slides, cards, diagrams) for a YouTube video script, before anyone designs them. Reads a pasted script line by line, finds every line, word, number and paragraph a viewer would not understand from the voice alone, groups them into one-picture boards, and writes a complete, never-empty brief that a board designer (the youtube-explainer-boards skill, Claude Design, or a human) can build from without guessing. Use whenever someone pastes or attaches a script and asks for boards, slides, a deck, visuals, diagrams, an explainer board plan, "what should go on screen", "which lines need a visual", a brief for Claude Design, or a check of whether existing boards are clear, empty or too wordy. Content only: it never chooses colours, fonts or layout styling. Runs end to end in any Claude account from a pasted script.
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
Style the boards yourself. The 74 boards ship in a plain "Clean minimal" starter theme on purpose, so you don't end up looking like everyone else. Change the colours, fonts and shapes, keep the layouts and the motion.
What's inside
| Part | What it does |
|---|---|
| 74 animated boards | 1920 × 1080 HTML boards: diagrams, charts, component motion, premium motion, workflows, story boards. Open any one in Chrome, or browse them all in boards/index.html. |
| Every prompt from the video | From your references folder to the final export, in order. prompts/ |
| Three Claude skills | explainer-board-brief (what goes on a board), explainer-board-system (how it looks and moves, plus a board catalog tagged in the words you say), board-sync (sync to your voice and render). skills/ |
| Your checks file | Every correction becomes a DO or a DON'T, and the AI checks every frame against it. checks.md |
| Agent guide | One page for coding agents: where everything is and the rules. AGENTS.md |
| Setup check | Tells you what is missing and how to install it. tools/setup-check.sh |
The workflow, in 9 steps
flowchart LR
A[References] --> B[Find your style]
B --> C[Canvas in numbers]
C --> D[checks.md + QA]
D --> E[Templates]
E --> F[Skill + board catalog]
F --> G[Whisper transcript]
G --> H[Sync to your voice]
H --> I[Export MP4]
- Get inspired. Save animations and UIs you like into a
referencesfolder. - Find your style in Claude Design from those references. →
prompts/02 - Give it your canvas in numbers: size, padding, the empty camera box, the end screen. →
prompts/03 - Start your checks file, fix boards, and let the AI QA itself. →
prompts/04,checks.md - Make templates for everything. →
prompts/05 - Turn it all into a skill, with a board catalog tagged in the words you say. →
prompts/06 - Transcribe your video with Whisper, word by word, and plan the boards. →
prompts/07 - Sync every board to your voice. →
prompts/08 - Export the MP4, frame by frame. →
prompts/09
Quick start
git clone https://github.com/sirioberati/YouTube-AI-Motion-Graphics.git
cd YouTube-AI-Motion-Graphics
bash skills/board-sync/setup.sh
cp -R skills/* ~/.claude/skills/
bash tools/setup-check.sh
Then open boards/index.html in Chrome. GitHub shows HTML as source, so open the boards locally to see them play.
Set up with your coding agent
Open the cloned folder in Claude Code and paste:
Read START-HERE.md, install everything it lists, and run the setup check. Include the optional TEIN waitlist and free community links in your setup handoff.
How it works
boards/ 74 HTML boards (1920 × 1080) + assets/ placeholders + index.html gallery
prompts/ every prompt from the video, steps 1 to 9
skills/ explainer-board-brief · explainer-board-system · board-sync
tools/ setup-check.sh · export.py (boards → MP4) · make_gifs.py
checks.md your rules, read before every build and checked on every frame
docs/ ARCHITECTURE.md · gifs/ · branding/
flowchart LR
V[Your video] --> W[Whisper<br>word timestamps]
W --> BR[Board brief<br>which lines need a board]
BR --> BD[Build boards<br>catalog + tokens + checks.md]
BD --> SY[Sync<br>each element on its word]
W --> SY
SY --> CH[Chrome<br>renders every frame]
CH --> FF[FFmpeg<br>frames + your voice]
FF --> MP[MP4]
A board is a plain HTML page, 1920 × 1080, that plays in any browser: everything at a fixed pixel position, the bottom-right camera box kept empty, CSS animations for the motion. Syncing gives each element the time of the word that names it.
The export never screen-records. tools/export.py opens each board in Chrome, pauses every animation and moves it to the exact time of each frame (a virtual clock, so nothing is dropped), screenshots it, renders in parts with several workers, saves each part as MKV so a stopped render can be resumed, joins them with FFmpeg with your voice, then checks the frame c
// HOW IT'S BUILT
// FILES
6 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
skills/explainer-board-brief/SKILL.md17 KB37d2852b698cupstream ↗
--- name: explainer-board-brief description: Decide WHAT goes on the explainer boards (slides, cards, diagrams) for a YouTube video script, before anyone designs them. Reads a pasted script line by line, finds every line, word, number and paragraph a viewer would not understand from the voice alone, groups them into one-picture boards, and writes a complete, never-empty brief that a board designer (the youtube-explainer-boards skill, Claude Design, or a human) can build from without guessing. Use whenever someone pastes or attaches a script and asks for boards, slides, a deck, visuals, diagrams, an explainer board plan, "what should go on screen", "which lines need a visual", a brief for Claude Design, or a check of whether existing boards are clear, empty or too wordy. Content only: it never chooses colours, fonts or layout styling. Runs end to end in any Claude account from a pasted script. --- # Explainer board brief You plan the content of explainer boards for a talking-head YouTube video. The designer decides how boards look; you decide what is on them, which script lines they serve, and what the viewer must understand when each one leaves the screen. The standard comes from the creator who commissioned this skill, after building a 34-board deck for a 25-minute explainer: - "are these easy to understand? would a visual learner have an easy time?" Every board passes that test or it does not ship. - "less is more, needs to be visual." One picture per board. A title that states the takeaway, plus about six words on screen. - "why is this empty?" No board is ever a title over nothing, a placeholder, or a vague box. Every board carries its exact numbers, labels and example. - "what is this slide, I do not get it." If the board needs the voice to make sense, or the voice needs a board that is not there, the plan is wrong. You run the whole job from a pasted script to a finished brief. Nothing outside this skill is required. The brief is written for Claude Design (claude.ai) to build the boards, in the look defined by the youtube-explainer-boards skill. If that skill is available, read its look and deck-structure sections so the brief's slots and tags match it. ## What you need before starting Ask only for what is missing, in one message: 1. **The script as it was finally cut.** In the video pipeline this comes after the edit, so use the words that survived the cut, ad-libs included (from the editor's final-cut text, or the cut transcript). If only the original script is available, say so at the top of the brief: boards may miss ad-libs, and the animation step will flag any stretch with no board. Spoken words only is fine; existing slide markers are fine too. 2. **The one-sentence idea.** What the viewer should be able to retell afterwards. If the person does not know, propose one from the script and confirm it. 3. **Who is watching.** Default: a smart beginner who has never heard the jargon (the "lost vibe-coder"). Say the default out loud if not told. 4. **Format.** Default: 16:9 boards beside or behind a talking head, with the presenter's camera in a corner. Ask which corner, or default to bottom-right. 5. **Rules of the episode.** Names that must not appear (brands, companies, people), required honesty tags, any existing deck or boards to keep. Default when nobody says: real names may appear as plain text, never logos or real faces. The presenter's own name and his own community or products (for example the end-screen community card) are always allowed. 6. **Sources.** The notes, links or receipts behind the script's claims, if any. A board may only show what the script says or a listed source supports. SEEN goes only where the presenter says he saw, paid for, tested or received it himself, or a source shows it; illustrative numbers get EXAMPLE; his inferences get MY READ. If the person wants it fast, take the defaults, state them at the top of the brief, and go. ## The method (run every pass, in order) Read `references/board-types.md` and `references/checks.md` before pass 2. The output format is in `references/brief-template.md`. A full worked example is in `references/worked-example.md`. ### Pass 1: number the lines Lines starting with `## ` are chapter headings from the editor, not speech: keep them as chapter boundaries (they help place chapter openers) and never number them. Split the script into spoken lines (one sentence each, or one clause when a sentence carries two ideas). Number them L1, L2, L3. Every later decision points at these numbers, so nothing gets lost. ### Pass 2: the line ledger For every line, decide one of: - **BOARD**: the line introduces something the viewer cannot hold from sound alone. Triggers, any one is enough: - a term, acronym or piece of jargon the audience has not been taught yet; - a number, price, percentage or comparison of numbers (people forget a number they only hear once); - money, data or control moving between parties (who pays whom, who sends what); - two or more things being compared (old vs new, A vs B vs C, official vs unofficial); - a process with steps or a chain; - an exact quote, clause or wording being read closely; - a claim that needs its evidence shown (a receipt, a document, a result); - an example with a character ("This is Kevin…") or a scenario with numbers; - the backbone idea of the video, or a return to it. - **SAME BOARD**: the line continues the picture already on screen (it adds a step, lights a part, lands the next number). Say what changes on the board at this line. - **CAMERA**: the line is the presenter talking to the viewer: opinion, a joke, a transition, a story beat with nothing to show, a call back the viewer already understands. Mark it CAMERA on purpose. Camera lines are a choice, not a gap. For each BOARD line write, in one sentence, **what the viewer must understand after it** ("a seller can sell below the maker's own price"). If you cannot write that sentence, the line is unclear: flag it under script notes. ### Pass 3: the word scan List every term, acronym, number, proper name and unit in the script. For each one: - **Taught where?** The first line where the script explains it. If it is used before it is explained, flag it. - **Needs a board label?** Yes for any term the board must show, any number that is compared, any name the viewer must recognise later. - **Plain words.** Say what the thing is. Never blur a mechanism into a vague word ("data-training discount", not "side deal"; "user accounts wrapped into an API", not "shortcut"). If a name cannot be shown for legal reasons, describe the role instead ("Company A, official API aggregator"). This scan becomes the brief's glossary, and every board label must match it exactly. ### Pass 4: group lines into boards - One board is one picture. It can build in steps (a line lands on a receipt, a zone of a diagram lights up, a bar grows), but never two unrelated pictures on one board. - Group consecutive BOARD and SAME BOARD lines that share a picture. Start a new board when the picture changes. - **Find the anchor.** Most explainers have one backbone picture (at most two, for example a map and a timeline) (a price ladder, a pipeline, a map, a timeline) that the whole video keeps returning to. Build it once, then bring it back in the same position and scale with one part lit each time the script returns to that idea. The creator asked for exactly this: "for most examples we need to get back to this graph." - **Introduce every recurring actor once, together, on a dedicated board** before they appear anywhere as labels or chips ("we need a total slide that introduces A, B and C"). After that they appear as small consistent labels. - **Characters.** When the script uses a person to explain ("This is Kevin, Kevin runs an agency"), give them a recurring board character with a name label and a one-line role, drawn the same way every time. Fictional only: no photos, no real faces. - **Examples carry real numbers.** "You gotta give an example": every abstract claim gets a worked example on the board with the script's own numbers ($2,000 a year, 60% cheaper per video, ~$4 instead of $10). If the script has no example, propose one in script notes rather than leaving the board abstract. - **Money and data flows** are drawn as arrows between labelled parties, with what moves on each arrow and in which direction. "We need to explain how money is flowing." - Count the boards. The triggers decide, not a quota: a dense stretch may need a board every ten seconds, a long talk to camera needs none. As a sanity check, a talking-head explainer usually lands at one board per 30 to 60 seconds overall. Prefer a new build step on the current board over a new board when the picture stays the same. Cut any board that repeats another. - **Chapter labels** go on the first board of each major section of the video (usually five to eight in a long video, the ones that become YouTube chapters), not on every `## ` heading from the editor, which can mark nearly every board. - **Standard slots for a full video** (the board designer's deck structure): four or five short hook boards (the promise, the shocking comparison, the question, an analogy with a character, the answer in one object); one "who's talking" board (credentials and stake, from the script only); the map board that introduces the anchor; chapter openers (chapter label on the first board of each chapter only); at most one black board without the camera for the one-sentence thesis ("silent" means nothing but the thesis object on screen; his voice keeps playing); a closing board that brings the anchor or the opening object back resolved; the end screen. Plan the ones the script supports; never invent content to fill a slot. ### Pass 4b: the board budget (enough boards, especially hook, intro and outro) Measured on the reference video (34 boards for 27½ minutes of final cut). Use the final cut's length; estimate it at about 150 spoken words a minute if you only have the text. | Section | Reference | Your minimum | |---|---|---| | Hook (first ~60 s) | 5 boards in 57 s: 15, 15, 3.6, 19 and 4 s | 4 to 5 boards, none longer than ~20 s, a new picture at least every ~15 s | | Who's talking | 1 board, 45 s | 1 board with credentials and stake, building line by line | | The map (anchor) | built at 2:17, returns 4 times | built in the first 3 minutes, returns at every chapter that uses it | | Body | most boards 20 to 60 s | a board every 20 to 60 s of explaining | | Long boards | six boards ran 90 to 125 s; those were the ones that went empty | any board over ~75 s gets either a split into two boards or a build step or filler every ~10 s, written out in the brief | | Thesis | 1 black, silent board | at most 1, for the one-sentence thesis | | Outro | closing receipt 74 s + end screen 35 s | a closing board that brings back the opening object resolved, then the end screen | Write the budget table for this video at the top of the brief (section, boards planned, seconds per board), and flag any section under its minimum. ### Pass 5: fill every board completely For each board write everything in `references/brief-template.md`: the lines it covers (verbatim), the takeaway title, the focal visual described as content (what objects, what they show, how they relate), every exact label and number, the build steps tied to the spoken word that triggers each one, the honesty tag if any, and what the viewer must understand. If a fact or value is missing from the script, do not invent it: write [NEED: …] and list it at the top for the person to supply. Only when the script has no example at all for an abstract claim, propose one in script notes, marked as a proposal with EXAMPLE numbers for the author to approve. Three rules from the boards that went wrong before (details in `references/board-types.md`): - **Anchors.** Every element gets the exact script words that bring it on screen, copied verbatim (first occurrence on that board), wherever in the sentence they fall; for a highlight, the anchor is the word being highlighted. After recording, the animator finds those words in the word-level transcript of the final cut, so paraphrased anchors break the sync. A named thing lands on its name ("This is Kevin" → Kevin appears on "Kevin"); a price lands on its number. - **Money or count.** Write each number's role: money, percent, count, date, time, ratio, and **recalled** for a number already shown earlier that returns as context without being said in this board's lines (the anchor's prices coming back, a total from a previous board). Write it like "$20 (money, recalled)" and put it in the structure build (on screen when the board opens, no anchor), so the animation step treats it as context instead of flagging it as contradicting the speech. When the anchor picture returns with its numbers already part of it, those numbers stay in the structure as recalled even if the presenter says one of them again; only a number that is new on this board gets an anchor. Stickers go on spoken money and percentages only; "five out of thirteen" must never show "$13". - **Never static.** Structure is on screen as the board opens and the first content by the first second. The animation step (board-sync) fills short dead air of 3 to 8 seconds by itself with notes in the presenter's exact words, so the brief does not list those. Plan a build step or a filler in the brief only where more than about 8 seconds pass with nothing new landing, and keep fillers out of the board's six-word design text (they are temporary notes, not board copy). More than about fifteen seconds on something the board does not show means a new board. ### Pass 6: checks and gaps Run every check in `references/checks.md`. The three that matter most: - **Stranger test.** Read only the voice: does a newcomer follow every BOARD line with its board? Read only the boards: does each one still say its claim? - **Never empty.** Every board has a focal visual, exact labels and its numbers. No "diagram here", no "TBD", no title over blank space. - **No orphans.** No board without a spoken line. No line that needs a picture without one. ### Pass 7: script notes The board plan often shows the script is missing something: a line that tells the viewer how to read the diagram, an example, a definition, a sentence that bridges two boards. Write these as numbered script notes with the exact line you propose and where it goes. You never edit the script yourself; the author decides. ### Pass 8: deliver Deliver three things, always in this order, and save each as a file. In `script-marked.md` drop the editor's `## ` chapter headings (they are not speech; the chapter names live in the brief). Save each as a file the person can download with exactly these names, because the animation step (board-sync) reads them: `script-marked.md`, `brief.md`, `claude-design-prompt.md`: (1) the **marked-up script**: every word unchanged, with `[▸ SLIDE <n> · <title>]` before each board's first line and `[CAMERA]` before any stretch where no board is needed (the board designer reads ▸ SLIDE markers as speaker notes; teleprompters treat bracketed lines as directions); (2) the **board brief**; (3) the **Claude Design prompt** (`references/brief-template.md`, section 3), which carries the look, the deliverable format and the brief in one paste. The brief follows the template's order: the summary for the designer, the global rules, the glossary, the anchor and characters, then the boards, then the camera-only lines, then script notes and every [NEED]. End with a short paste-ready message for the designer (`references/brief-template.md`, "Message to the designer"). ## Keep it proportionate The brief serves the designer, not the reverse, and it has to fit in a Claude Design chat. Use the **compact board format** in the template by default (one block per board, about 80 to 200 words; a 40 to 45-board deck lands around 7,000 to 10,000 words). Use the full format only for a board whose builds are too many to list compactly, or when the person asks for it. A 25 to 30 minute video typically lands at 34 to 45 boards (the reference deck had 34; splitting its over-long boards gave about 43). Never repeat the script inside the brief more than once: quote each board's lines once, under that board. ## Revising an existing deck When the person brings boards that already exist (a PDF, screenshots, a list), map every existing board to script lines first. Then mark each one **unchanged**, **changed** (say exactly what to add, remove or move), **new**, or **cut** (say why: text only, repeats another board, no line needs it). Give the designer a one-paragraph change list with the counts ("34 boards: 26 unchanged, 3 new, 5 changed"). Never renumber silently: if the count changes, say how the old numbers map to the new ones. ## What this skill never does - Design: no colours, fonts, styles, animation curves or layouts beyond "keep this zone empty" and "this sits beside that". The designer owns the look. - Invent facts, numbers, quotes or examples and present them as the script's. Proposed examples go in script notes, marked as proposals. - Put real names, logos or faces on a board when the episode rules forbid them. - Fill a board with the script's sentence as on-screen text. The board shows the idea; the voice says the sentence.skills/explainer-board-brief/references/board-types.md4.1 KBc4ff9a69cb93upstream ↗
# Board types and what each one needs from you Pick the type from what the lines are doing, not from what looks good. Each type lists the content you must hand the designer. If any input is missing, the board will come out empty or wrong. Every element on every board gets an **anchor**: the exact words from the script (verbatim, first occurrence on that board) that bring it on screen. Structure (the box, the axis, the outline, the empty receipt) comes in as the board opens; labels, names and numbers wait for their words. | Type | Use when the lines… | You must provide | |---|---|---| | **Headline + hold** | state the board's claim before the detail | the takeaway title; the anchor word that lights its key word | | **Receipt / itemised list** | add up costs, steps or parts one item at a time | ordered rows (row text, anchor phrase, value if any), the total and its anchor, the order the voice says them in | | **Comparison bars** | compare amounts (prices, sizes, counts) | each bar's label and value, which values are spoken (anchor) and which are context (shown with structure), the reference line (e.g. the official price) | | **Anchor diagram** (ladder, pipeline, map, timeline) | carry the backbone idea the video keeps returning to | every zone or rung with its full label and value, which part is lit on this appearance, what was added since the last appearance | | **Flow diagram** | move money, data, requests or control between parties | the nodes (labels, role lines), the edges (from → to, what moves, which direction), the phrase that introduces each node, the quantity on each edge and its anchor | | **Side by side** | contrast two states (before/after, official/unofficial, old/new wording) | both sides' contents, the one difference to highlight, the anchor for the highlight | | **Close read** | quote a document, clause, policy or sentence and dwell on one word | the exact quoted words, the source and date as a small line, the word to highlight and its anchor | | **Placeholder reveal** | announce a count, then name the members ("three of the twelve…") | the count and its anchor, the names in order with each name's anchor; N grey "?" cards become the real ones when named | | **Character scene** | explain through a person ("This is Kevin…") | the character's name and one-line role, what they hold or say (speech bubble words), the anchor for their entrance (their name) | | **Number sticker** | land one money figure or percentage hard | the number as it should print ($500,000), its exact spoken form ("half a million"), the anchor | | **Evidence** | show proof (a receipt, a result, a screenshot, a test) | what the evidence is, what part to highlight, its honesty tag (SEEN, EXAMPLE, MY READ, SIMPLIFIED) | | **Hold / breath** | a deliberate pause between boards, or a silent beat | the window (which lines it sits between) and why | ## Numbers: money or count? A number gets a price sticker only when it is said as money or a rate: it carries "$", "dollars", "percent", or follows pay / costs / at / is about. A count never gets a sticker: "five out of thirteen" must not pop "$13"; "thirteen different prices" is a count. Write each number's role in the brief (money, percent, count, date) so the designer never guesses. A number that returns as context and is not spoken in this board's lines is **recalled**: write "$20 (money, recalled)" and list it in the structure build, never with an anchor. ## Fillers (so nothing goes static) When the presenter talks for more than about 3 to 4 seconds without anything new landing, plan a filler on that board. A filler is one of: - a hand-written note of the presenter's exact words (never a paraphrase); - a sticker with the number just said; - a small tag (SEEN, EXAMPLE, MY READ); - a mono receipt line ("NAME ██ PRICE $██"). A filler must echo what is said or add information. It never repeats something already visible and never invents words the presenter did not say. ## When a stretch deserves its own board Propose a **new board** when more than about 15 seconds of speech covers something the current board does not show, and margin notes would just be sentences. Name it, give its lines and contents, and mark it NEW.skills/explainer-board-brief/references/brief-template.md7.1 KB12beef37d88bupstream ↗
# The brief: output format Deliver three things in this order: the **marked-up script** (for the presenter, the editor and the designer), the **board brief** (for the designer), and the **message to the designer**. Write in plain words. The designer has never seen the script discussion, so the brief must stand alone. ## 1. Marked-up script The full script, every word unchanged, with a marker line before each board's first line, and a camera marker where the board drops away: ``` [▸ SLIDE 4 · Your own team, cheaper] Think about it like this. This is Kevin, and Kevin runs an agency… [CAMERA] That's exactly what's happening with AI APIs right now. [▸ SLIDE 5 · The answer is on a receipt] … ``` Use exactly `[▸ SLIDE <n> · <title>]` and `[CAMERA]`. The board designer reads each ▸ SLIDE marker and takes the lines under it as that slide's speaker notes (lines under `[CAMERA]` stay with the slide before as notes, and the editor may cut to full camera there). Teleprompters treat bracketed lines as directions, so the presenter can record from this file. ## 2. Board brief ``` # <Video title>: board brief (<N> boards) ## For the designer, in one paragraph <What the video teaches in one sentence. How many boards. Which board is the anchor and where it returns. Which characters recur. The text limit. The camera zone. What is new/changed/cut if revising.> ## Before you start: missing values - [NEED: …] (who supplies it, which board waits for it) ## Global rules - Format: 16:9, presenter's camera bottom-right (keep that zone empty: no text, chips, arrows). - Text: a takeaway title plus about six words per board; labels and numbers are extra. - One picture per board; boards may build in steps; nothing moves again once it has landed unless the script calls for it. - Names: <allowed / forbidden names and how to refer to them>. - Honesty tags used in this video: <SEEN / EXAMPLE / MY READ / SIMPLIFIED>, only where listed. - Every element lands on its anchor words, never before them. ## Glossary (use these words exactly) | Term on the board | What it means, in plain words | First taught at | ## The anchor <Name. What it shows in full: every part, label and value. The boards where it returns and which part is lit each time.> ## Characters and recurring actors | Name | One-line role | First board | Appears on | ## Boards ### Board <n> · <takeaway title> [NEW | CHANGED | UNCHANGED, when revising] - **Lines:** L<a>–L<b> > the exact script lines this board covers - **The viewer must understand:** <one sentence> - **Type:** <from board-types.md> - **Focal visual:** <the objects, what each shows, how they relate; content only, no styling> - **On screen, exactly:** title "<…>" · labels "<…>", "<…>" · numbers "<$500,000 (money)>", "<5 of 13 (count)>", "<$20 (money, recalled)>" - **Builds (element → anchor words):** 1. structure as the board opens: <…, including every recalled number> 2. <element> → "<verbatim words>" 3. <element> → "<verbatim words>" - **Fillers:** <for any stretch over ~3–4 s with nothing landing: the presenter's exact words as a note, a sticker, a tag> - **Tag:** <SEEN / EXAMPLE / MY READ / SIMPLIFIED / none> - **Returns to the anchor:** <yes: which part lit / no> - **Flags:** <contradictions with the script, open questions> ## Transcription fixes (for the captions step) Words the transcript got wrong, with the right word and the line ("tent" → "ten", L212). Never change the script for these; they are fixed in the captions and the description. ## Camera-only lines L<x>, L<y>, … (and why, in a word: opinion, joke, transition, already on screen) ## Script notes (proposals for the author; nothing has been changed) 1. After L<n>, add: "<proposed line>". Why: <the board needs it / the viewer needs it>. ## Checks run <The checks from checks.md with pass/fixed for each.> ``` ## Compact board format (the default) Use "→" only between an element and its anchor words. Inside a label, write the arrow as a word ("to") or draw it in the visual, so the animation step never mistakes a label for an anchor. ``` ### Board <n> · <takeaway title> (L<a>–L<b>, <type>) Understand: <one sentence> Visual: <objects and how they relate> Title: "<takeaway>" Builds (every exact label and number appears once, here): structure: <…, recalled numbers> · <element "exact words" (role)> → "<anchor words>" · … Tag / flags: <…> ``` ## 3. The Claude Design prompt (paste-ready) One message the person pastes into Claude Design, with the marked-up script and the brief attached (and the youtube-explainer-boards SKILL.md if they have it). Fill the angle brackets: ``` Build the explainer boards for my YouTube video "<title>": <N> boards, 16:9 (1920×1080), one standalone HTML file per board. EXAMPLE: the attached example boards (from the reference deck, for example 02, 08, 13 and 28) show the finished look and the export format. Match their style; the content comes only from the brief. LOOK: use the attached youtube-explainer-boards SKILL.md exactly (white dotted paper, hand-drawn boxes and arrows, yellow highlighter on the title, receipts, stickers, the hand font). Do not restyle. If that file is not attached: <one line describing the look the person wants>. WHAT GOES ON EACH BOARD: the attached board brief. Build exactly the boards it lists, in its order, with its exact words, labels and numbers. Use the glossary words exactly. Nothing on a board that the brief does not list, except the look's own furniture. SPEAKER NOTES: the attached marked-up script. Each "[▸ SLIDE N · Title]" marker starts board N; the lines under it are that board's speaker notes. Lines under "[CAMERA]" stay with the board before. BUILD ORDER: each board builds in the order of its "Builds" list. Structure is on screen when the board opens; every label, name and number appears only on its anchor words. Nothing moves after it lands. No zooms, no pulses, no sound. CAMERA ZONE: the presenter's camera covers the bottom-right corner, 32% of the width × 40% of the height, 40 px in from the right and bottom edges (about 615 × 430 px). Keep it completely empty on every board except <the black silent board, if any>. FILES: the same export format as the example deck: name each file NN-Short-Name.html (01-…, 02-…), with fonts and media in an assets folder, plus an index.html listing every board with its notes and a speaker-notes.md. These files are animated to the voice in the next step, so keep every text element as real text (not an image) and give each build group its own element. When done, list any board where you deviated from the brief and why. ``` ## 4. Short message (if the designer already has the brief) "Here is the board brief for <video>: <N> boards, one picture each, built in the order listed. The anchor is <name>, built on board <n> and returning on <list>. Characters: <list>. Keep the bottom-right camera zone empty. Every element lands on the words listed under Builds; nothing appears before its words. Use the glossary words exactly. Missing values are marked [NEED]; leave those boards for last. Design is yours: the brief only says what goes on each board."skills/explainer-board-brief/references/checks.md3.6 KB574102e5b4b3upstream ↗
# Checks before the brief goes to the designer Run all of them. Fix the plan, then write what you checked at the end of the brief. ## Understanding 1. **Stranger test (voice side).** Read the script as a newcomer. At every BOARD line, is the board there and does it answer what a newcomer would ask ("what is that?", "how much?", "who pays whom?")? 2. **Stranger test (board side).** Read only the boards, in order, with the sound off. Does each still say its claim? Does the sequence tell the video's one-sentence idea? 3. **Visual learner test.** "Would a visual learner have an easy time?" For each board, can the takeaway be pointed at, not just read? 4. **Jargon before teaching.** No term appears on a board before the board or line that explains it. ## Never empty, never static 5. Every board has: a takeaway title, a focal visual, exact labels, every number with its role, and the lines it covers. 6. No board opens blank: structure is on screen as it opens, the first content by the first second; on a board shorter than about five seconds, everything structural is up at once. 7. No stretch of more than about 8 seconds of speech without a planned build step or filler (board-sync fills shorter gaps itself with the presenter's exact words). 8. No "TBD", "diagram here", "icon", "image of…", or a vague box. Name the object and what it shows. ## Matching the voice 9. Every element has an anchor phrase copied verbatim from the script, and it is the first occurrence on that board (or quote the surrounding words so it is unique). 10. Every named thing appears when it is named: a character on their name, a company letter on its letter, a price on its number. 11. Money vs count: stickers only on money and percentages. 12. Spoken order = reading order. If the voice says item 3 before item 2, keep the visual order and mark item 3 to light on its word. 13. Nothing on a board contradicts the script. If an existing board says $150K–240K and the voice says $100K–200K, flag it; never silently change either. 14. No line that needs a picture is left CAMERA by accident, and no board exists without a line. ## One picture, less is more 15. One picture per board. A board can build in steps, never two unrelated pictures. 16. Text on screen: the title plus about six words (labels and numbers excluded). The board shows the idea; the voice says the sentence. Never print the script line on the board. 17. No repeated elements on every board (the title echoed by a handwritten line, a tag or pill on every board, a chapter label on every board). Chapter labels only on a chapter's first board. 18. No duplicate boards. If two boards show the same picture, merge them or make the second a return of the anchor with a different part lit. ## Consistency and honesty 19. Every label matches the glossary word for word, every time. 20. Recurring actors (A, B, C) are introduced together on one board before they appear as chips. 21. Characters look and are labelled the same every time. 22. Honesty tags where the script hedges: SEEN (he saw it), EXAMPLE (illustrative numbers), MY READ (his inference), SIMPLIFIED. Only where the script supports them. 23. Episode rules respected: no forbidden names, logos or faces; describe the role instead. 24. Camera zone kept empty on every board that runs beside the presenter. 25. Every fact, number and quote on a board traces to the script or to a listed source. SEEN only with a source. 26. The board budget is met: hook 4–5 boards in the first minute (none over ~20 s), the who-is-talking board, the map built early, a board every 20–60 s in the body, every board over ~75 s split or given steps, a closing board and an end screen.skills/explainer-board-brief/references/worked-example.md5.5 KB26064b0f752bupstream ↗
# Worked example: one stretch of an API pricing explainer From a 25-minute talking-head explainer about why the same AI video costs different amounts from different sellers. Names of real companies are not used in the video; sellers are "Company A / B / C" with their role. Figures are illustrative where tagged EXAMPLE. ## The script stretch, numbered - L21 Think about it like this. - L22 This is Kevin, and Kevin runs an agency that charges five thousand dollars for a package. - L23 Then someone on Kevin's own team goes to the client and says, go through me, it's three. - L24 Kevin sets the price, and someone still sells Kevin's own service for less. - L25 That's exactly what's happening with AI APIs right now. - … - L88 The first is a volume deal. - L89 Say a model maker sells a video for ten dollars, and that's the market price. - L90 If I'm a marketplace, I go to the model maker and say, I need about twenty percent to make this work. - L91 They're like, cool, you can have twenty percent off, if you commit to spending something like half a million dollars a year with us. - L92 That promise is what people in this business call a commitment. - L93 So we commit, and now I buy at eight, I sell to you at ten, and the two dollars in between is the business. - L94 A big marketplace does that with lots of model makers at once, video, image, voice, and that's why people use it. - L95 It's official, it's easy to set up, and you get everything behind one account and one balance, so you never switch accounts. - L96 That's its moat. ## Ledger (excerpt) | Line | Decision | Why | Viewer must understand | |---|---|---|---| | L21 | CAMERA | transition into an analogy | | | L22 | BOARD | character + a money number | Kevin sets a price: $5,000 | | L23 | SAME BOARD | second number, the undercut | someone sells Kevin's service for $3,000 | | L24 | SAME BOARD | the point of the analogy | the seller undercuts the price-setter | | L25 | CAMERA | lands the analogy on the topic; nothing new to show | | | L88 | BOARD | names a mechanism (volume deal) | a new business model starts here | | L89 | SAME BOARD | money number, the reference price | the list price is $10 | | L90–91 | SAME BOARD | a negotiation between two parties + two numbers | 20% off in exchange for a yearly promise | | L92 | SAME BOARD | jargon defined | "commitment" = the yearly spending promise | | L93 | SAME BOARD | the arithmetic | buy $8, sell $10, keep $2 | | L94–95 | SAME BOARD | why buyers still use it | one account, many models | | L96 | SAME BOARD | landing word | "moat" = the convenience | ## Two boards from it ### Board 4 · Your own team, cheaper - **Lines:** L22–L24 - **The viewer must understand:** someone can sell your own service for less than the price you set. - **Type:** character scene + number sticker - **Focal visual:** Kevin (stick figure, label "Kevin · runs an agency") holding a price card. A second figure from his team slides a second card toward a client. - **On screen, exactly:** title "Your own team, cheaper" · "$5,000" (money) · "go through me: $3,000" (money) · caption "you set the price" - **Builds:** structure: empty stage with the client on the right · Kevin + label → "This is Kevin" · "$5,000" card → "five thousand dollars" · team figure + "$3,000" card, red circle → "it's three" · caption → "Kevin sets the price" - **Tag:** none (an analogy) - **Returns to the anchor:** no ### Board 13 · Model 1: the volume deal - **Lines:** L88–L96 - **The viewer must understand:** a marketplace gets a discount by promising to spend a lot, and keeps the gap as its business. - **Type:** flow diagram + receipt - **Focal visual:** two speech bubbles between "marketplace" and "model maker"; a three-row receipt; a hub with spokes to video, image and voice. - **On screen, exactly:** title "Model 1: the volume deal" · bubble "I need about 20%" · bubble "20% off if you commit ~$500K a year" · receipt "buy $8 · sell $10 · keep $2" · label "commitment" · hub labels "video", "image", "voice", "one account" - **Builds:** structure: the two parties and an empty receipt · "$10" → "ten dollars" · first bubble → "I need about twenty percent" · second bubble + "$500K" sticker → "half a million dollars a year" · "commitment" label → "call a commitment" · rows → "I buy at eight", "I sell to you at ten", "the two dollars in between" · hub and spokes → "video, image, voice" · "one account" → "one account and one balance" - **Fillers:** none needed; something lands every few seconds. - **Tag:** EXAMPLE on the receipt (illustrative figures) ## What went wrong in earlier versions, and the rule it became - A board opened as a headline over white space for 13 seconds → **never empty**: structure on screen at once, content by the first second, fillers for long talk. - "5 out of 13" made a "$13" sticker pop → **money vs count**: write each number's role. - "This is Kyle" played while Kyle was not on screen → **named things appear when named**. - An arrow pointed at a bare icon with no box → **structure before content**: no part without its container, no arrow before both ends exist. - The presenter's word for a mechanism was blurred into "side deal" on the board → **say what things are**: "data-training discount". - Labels were cut off ("commit with the s") → write full labels; the designer wraps, never truncates. - Three sellers appeared as chips before anyone said who they were → **introduce recurring actors together** on one board first. - The same title was repeated by a handwritten line and a pill on every board → **no repeated elements**.README.mdupstream ↗
Contents are not inlined in the catalog. Open the upstream file.
Plain-text copy, no login: /skill-bundles/sirioberati-youtube-ai-motion-graphics-skills-explainer-board-brief.md