Studio Video Creator

This skill produces a single, well-defined artifact: a 60–90 second concept film. Voiceover-driven, prototype-grounded, insight-led, aspirational close.

The film is used at the moment a studio concept becomes legible — post-research, pre-build — to frame the idea for an internal team, an LP, a launch audience, or the founder themselves. It is not marketing collateral. It is a thinking artifact rendered as video.

Philosophy: Guide like software, not chat

The user wants a tool, not a conversation. Behave accordingly:

Decide silently where you can. Variant selection, structural choices, design defaults — make the call, then name it briefly. Do not present menus when you can present a recommendation.
Ask only what you genuinely need. Hard cap: three questions in the entire intake phase. Each must change the output materially. If a question can be answered from the source material, do not ask it.
Gate every phase. Do not advance to script without a legible concept summary. Do not render without a chosen voice. State the gate; don't quiz the user about it.
Show progress. Each phase ends with a short status line ("Concept locked → moving to script") so the user always knows where they are.
Withhold options. Curation is the value-add. The voice list is 8, not 100. The framework variants are 3, not 12. If the user wants more, they ask.
Recover gracefully. Missing prototype? Suggest the minimum surface needed. Vague brief? Restate what you can extract and ask the user to confirm or correct.

Execution context — interactive sessions only

This skill requires an interactive session where the user can respond between phases. The workflow is a series of human-judgment gates; without the human, the output can look complete before the important decisions have actually been judged.

If running inside an autonomous task, batch prompt, single-shot delegation, scheduled job, or any context where the user cannot provide input between phases: STOP after Phase 1, present the brief, and tell the user to run the skill in an interactive Claude Code session to continue.
The skill's value is in its gated process. Skipping gates usually produces output that appears finished before the concept, design, voice, or hook has been properly reviewed.
Do not delegate this skill to an autonomous agent. Do not pre-answer the gates in a single prompt. Each gate exists because the downstream phase depends on human judgment at that gate.

When to use this skill

Use when:

The user has source material (brief, research, prototype, deck, URL, PDF) for a product concept or studio project AND wants a video
The user is at the "this needs to become legible" moment in a concept's life
The video's purpose is framing/pitching/introducing — not marketing or tutorial
The target length is short (60–120s)

Do not use when:

The user wants a long-form video, tutorial, course, ad creative, social post, talking-head, or recorded demo
The user has no source material and just wants to brainstorm — that's a different mode
The user explicitly wants something other than the concept-film genre

If unsure, ask one disambiguating question and let the user steer.

What you produce

By the end of a complete run, the user has, in their project directory:

<project>/
├── brief.md           — interpreted concept, audience, vision statement
├── script.md          — final script with section timing
├── motion-board.md    — beat-by-beat visual causality plan
├── design.md          — chosen visual/motion language, brand interpretation, cover strategy
├── voice.json         — selected ElevenLabs voice ID + audition notes
├── renderer.json      — selected local renderer (`remotion` by default, `hyperframes` when warranted)
├── remotion/ or hyperframes/
│                      — renderer project, scenes filled with project content
│   ├── package.json
│   ├── data/ or src/
│   └── public/
└── out/
    ├── design-thumbnail.png — Phase 4 title-frame / style-frame artifact
    ├── cover-frame.png — actual frame-0 poster / preview image from the hook
    ├── hook.mp4       — first 10–15s rendered (round-one deliverable)
    └── final.mp4      — full film (rendered on explicit request)

The design thumbnail is the aesthetic iteration unit. The motion board is the explainer-quality gate. The cover frame is the silent first-impression check. The hook render is the film iteration unit. The full render is the publication unit.

Workflow

The workflow has seven numbered phases plus a Phase 3B motion-board gate. Each phase is a gate. Do not skip ahead.

Phase 1 — Intake & legibility gate

Receive the user's source material. Inputs may include:

A brief, research doc, or memo (PDF, MD, DOCX, plaintext)
A prototype reference (Figma link, deployed URL, screenshots)
A deck (PPTX, PDF, Keynote export)
A website URL
A loose description in the user's own words

Steps:

Read everything provided. Use web_fetch for URLs, the file-reading skill conventions for documents, image analysis for screenshots.
Produce a two-sentence concept summary — what it is and why it matters. Show this to the user.
Gate: Ask the user to confirm or correct the summary. Do not proceed until the concept is legible in two sentences. If you cannot summarize the concept in two sentences from the inputs, the video will fail. Tell the user this directly and ask for the missing piece.
State what minimum prototype surface the video will need ("To do this concept justice, the prototype should show: the home/landing view, the core ontology screen, and the action moment"). If the prototype provided already covers this, say so. If not, flag the gap — the user can either provide it now or accept that the film will use schematic representations.

End Phase 1 with: Concept locked. → Phase 2: clarifying questions.

HARD STOP. Present the two-sentence summary and wait for user confirmation. Do not proceed to Phase 2 until the user explicitly confirms or corrects. Silence is not confirmation.

Read references/intake-checklist.md for the full checklist of what to extract from source materials and how to handle each input type.

Phase 2 — Three sharp questions (and not one more)

Ask at most three questions. Skip any whose answer is already in the source material.

The three slots:

Audience + emotional state. Who watches this, and what state are they in when the film starts? (Examples: "Skeptical LP who's seen 20 pitches this week." "Internal team that needs to believe.") This determines tone.
The single insight. If the viewer remembers one thing 10 minutes after watching, what is it? Force the answer into one sentence. This becomes the spine.
The vision statement. "In a world where… / We believe… / What if…" — one sentence the film can land on. If the user is unsure, propose two options from the source material.

If the source material answers any of these, skip the question and state your inference: "I'm reading the audience as institutional LPs. Correct me if not."

End Phase 2 with: Brief assembled. → Phase 3: structure and script.

HARD STOP. Present the assembled brief (audience, insight, vision statement) and wait for confirmation.

Phase 3 — Variant selection and script

You have three structural variants. Pick one silently based on the brief.

Customer-led — opens on a person living the problem. Use when the concept is grounded in a human moment (mobile medical care, education, housing, services). The viewer enters through empathy.

Insight-led — opens on the idea itself. "What if X." Use when the concept's power is conceptual and the human moment is harder to dramatize (B2B tools, infrastructure, platform plays). The viewer enters through curiosity.

Demo-led — opens cold on the product, voiceover catches up. Use when the product is visually striking and se

studio-video-creator