media-notebooklm-slide-prompt

media-notebooklm-slide-prompt — NotebookLM slide deck + Nano Banana image prompt builder

One-line summary

Enter the session body markdown once, and a slide deck prompt to paste into NotebookLM Studio와 Nano Banana (Gemini 3 Pro Image) image prompts for each slide's visualsare produced simultaneously.

Boundary between this skill and doc-pptx

Even for the same "presentation materials," the outputs differ. Before invoking, distinguish them using the table below.

Category	media-notebooklm-slide-prompt	doc-pptx
Output	NotebookLM source organization + Studio Prompt text + per-slide image prompts	Editable .pptx file (pptxgenjs code)
Where to view the result	NotebookLM Studio generates the slides	Opens directly in PowerPoint/Keynote
Input	Lecture/class body MD	Report/planning content + design requirements
When to use	"I need a prompt to make presentation materials with NotebookLM"	"I need a PPT file that opens right now"

Summary: The builder that creates the materials to put into NotebookLM (source, script, structure, image prompts)is this skill. If you actually need a .pptx file, call moai-officer:doc-pptx. "Export it as a PPT file" is moai-officer:doc-pptx, and "Make me a NotebookLM slide prompt" is this skill.

Trigger keywords

When the following expressions appear, invoke this skill first:

• "NotebookLM prompt", "NotebookLM slide", "NotebookLM deck"
• "lecture slide prompt", "lecture slide materials", "session slide prompt"
• "Detailed Deck", "Presenter Slides", "presenter slides", "detailed deck"
• "slide image prompt", "Nano Banana slide", "Nano Banana slide"
• "Convert MD into a prompt to put into NotebookLM"
• "Make me a prompt to turn the body into slides"
• "A prompt to make the entire lecture presentation materials"

Input

Required

• Session body MD — .md file or attach the body directly. An Introduction → Core → Demo → Wrap-up 4-block structure is ideal.

Optional (supplemented via interview)

• Target slide count (e.g., 12 slides / 15 slides / 20 slides)
• Format preference (Detailed Deck vs Presenter Slides)
• Length preference (short / default / long)
• Output language (default: Korean)
• Audience/tone (e.g., "beginner–intermediate developers + students, candid hands-on retrospective style")
• Visualization concept (e.g., "minimal diagram + 1-cut cinematic cover")
• Brand constraints (font, color, logo — optional)

Workflow (Phase 1-4)

Phase 1 — Intake & Interview

Check whether the body MD is attached. If not, receive at least a one-line summary of the topic and key points. Group missing optional items into one round of AskUserQuestion (within 3-4 questions) and ask:

1. What is the target number of slides?
2. Detailed Deck vs Presenter Slides — which one?
3. What is the tone of the visual materials? — references/slide-style-library.md의 49 styles × 8 categories 1 category matching the presentation keywords from the libraryAfter automatically recommending 2, present 1 of them to the userPresent 3 style candidates (e.g., for an "AI·future·innovation" presentation, candidates 7·22·43 from category E)
4. Are there any slides or messages that must be covered?

Phase 2 — NotebookLM slide prompt generation

NotebookLM Studio의 Official 4 axesMap exactly to:

축	Mapping
Format	Detailed Deck(for reading·full text) or Presenter Slides(for presenting·key talking points) — based on interview answers
Length	short / default / long — automatic mapping based on target slide count (≤10 = short, 11-18 = default, ≥19 = long)
Output language	Default Korean. Specify if the audience is multilingual
Prompt body	6-block template below

NotebookLM Prompt body 6-block template

[1] 청중 & 사전 지식
   - 이 슬라이드를 볼 사람과 그들의 사전 지식 수준을 명시
   - 예: "클로드 코드 입문~중급 개발자 + 학생·취준생. 에이전틱 코딩 용어는 처음일 수 있음."

[2] 1순위 메시지 (The Single Takeaway)
   - 발표가 끝났을 때 청중이 한 줄로 기억할 메시지
   - 예: "하네스가 있어야 에이전트가 통제된 작품을 만든다."

[3] 슬라이드 구조 지침
   - 표지·도입·핵심·데모·정리·Q&A·다음 단계 — 어떤 흐름으로 배치할지
   - 매수 가이드 (예: "총 15장 ± 2장. 핵심 개념은 1슬라이드 1메시지.")

[4] 톤 & 스타일
   - 어조(예: "솔직한 실무 회고체, 클리셰 금지")
   - 시각 톤(예: "다이어그램 위주, 텍스트 60% 이내, 표지·섹션 구분만 풀이미지")

[5] 강조 슬라이드 (Must-Have Slides)
   - 절대 빠지면 안 되는 슬라이드와 그 슬라이드가 답해야 하는 질문

[6] 금지 / 제약
   - 다루지 말 것 · 피해야 할 표현 · 출처 표기 규칙
   - 예: "혁신적·차세대·재정의하는 같은 클리셰 금지. 코드 인용은 모노스페이스."
   - 슬라이드 제목 작성 규칙 (필수): 대시 대비 헤드라인("X — Y" 대시 분할 구조) 금지, 조사·체언 종결 조각문(조사/체언으로 끝나는 파편문) 금지, "A에서 B로" 전환 공식 금지. 제목은 완결된 명사구 또는 서술어를 갖춘 완전문으로 작성한다. (이 규칙은 본 템플릿이 생성하는 다운스트림 카피가 게이트 스킬 없이 바로 쓰이기 때문에 템플릿 자체에 새긴다.)

Output format (NotebookLM part)

## NotebookLM Studio 입력값

- Format: Presenter Slides
- Length: default
- Output language: 한국어

## NotebookLM Prompt (이 블록 전체를 Studio의 'Prompt' 칸에 그대로 붙여 넣으세요)

```
[1] 청중 & 사전 지식
…

[2] 1순위 메시지
…

[3] 슬라이드 구조 지침
…

[4] 톤 & 스타일
…

[5] 강조 슬라이드
- 슬라이드 3: "왜 하네스가 필요한가?" — …
- 슬라이드 7: "TRUST-5 다섯 가지 원칙" — …
…

[6] 금지 / 제약
…
```

Cautions when using official sources — NotebookLM uses only sources uploaded to the notebook. Upload the body MD to the notebook as a source first, then enter the above Prompt into the slide deck Prompt field in Studio. Do not paste the entire body into the Prompt field.

Phase 3 — Nano Banana image prompt generation (per slide)

Extract key slides from the body MD. Default extraction targets:

• Cover (1 slide)
• Section divider slides (usually 2-3 slides)
• Must-Have slides (emphasis slides received in the interview)
• Summary·conclusion (1 slide)

총 5-8 slidesis typical, and for each slide DeepMind official 5-Component structuregenerate a prompt filled in.

5-Component (required)

Component	Definition	Writing guide
Style	Illustration/photo/watercolor/cinematic/flat diagram, etc.	references/slide-style-library.md의 49 style catalogSelect 1 from and use the English prompt keywords as-is — e.g., "minimal isometric diagram (#39)", "cinematic space SF (#22)"
Subject	Who/what — specifically including appearance, clothing, and pose	Personify or objectify abstract concepts — e.g., "the agent is a small robot wearing a helmet"
Setting	Background·time of day·mood	Semantically connected to the lecture topic — e.g., "a dark workspace lit by terminal light"
Action	What action is taking place	Even for a still cut, specify the moment of a scene — "the moment a hand comes down on the keyboard"
Composition	Angle, framing, aspect ratio	For slides, 16:9, for the cover, wide cinematic, for section dividers, center-weighted

Additional options (optional)

• Text rendering — specify the exact wording in double quotes. E.g., the words "하네스 엔지니어링" rendered in a bold sans-serif font
• Diagram mode — a clean isometric diagram of an orchestrator-worker pattern, labeled arrows specify information visualization, such as
• Aspect ratio / resolution — 16:9 widescreen, upscale to 2K recommended (default 16:9 for lecture slides)
• Series consistency — enforce the same Style + palette + lightingacross all visual materials of a lecture. When characters appear, assign fixed names (e.g., the agent named "Trust-5")
• Safety — for human faces, virtual characters, silhouettes, or back shots are recommended. Depiction of real people is prohibited.

Output format (image portion)

## 슬라이드별 나노바나나 이미지 프롬프트

### 슬라이드 1 — 표지: "클로드 코드로 시작하는 실전 에이전틱 코딩"

Style: cinematic minimal, dark teal palette with warm amber accents, soft volumetric light
Subject: a small humanoid agent figure made of brushed metal, mid-stride, holding a glowing harness in one hand
Setting: a dimly lit late-night workshop with terminal-blue holographic interface panels floating around
Action: the agent fastening the harness around a translucent code construct that hovers above the desk
Composition: 16:9 widescreen, low-angle hero shot, center-weighted, generous negative space on the right for the title overlay
Text render: the words "하네스 엔지니어링" in a bold sans-serif font, lower-right, no other on-image text
Aspect ratio: 16:9, upscale to 2K
Consistency tag: series="harness-lecture", palette="teal-amber-dim", lighting="volumetric-warm"

### 슬라이드 3 — "왜 하네스가 필요한가?"
…

### (이하 슬라이드별 반복)

Phase 4 — Output integration & handoff

Deliver as a single bundled markdown file:

• Recommended filename: prompts/S{n}-notebooklm-prompt.md (based on the user's project path)
• Structure:
  1. Meta (session, audience, slide count, format)
  2. Part A — NotebookLM Studio input values + Prompt body
  3. Part B — Per-slide Nano Banana image prompts (5–8 slides)
  4. Usage procedure (upload the body to NotebookLM → paste the Prompt → generate → generate cover and key slide images separately with Gemini/Nano Banana Pro, then replace them in NotebookLM via revise)

Just before the chain ends, moai-coworker:ai-slop-reviewerto remove clichés and translationese.

Usage examples

Example 1 — First call (with interview)

사용자: "S0 도입 본문을 NotebookLM 슬라이드 프롬프트로 만들어줘"

→ Read the body MD and ask the four items—slide count, format, visual tone, and emphasized slides—in one round of AskUserQuestion.
→ After receiving the answers, deliver Part A + Part B as a single markdown file.

Example 2 — Specify slide count and tone directly

사용자: "본문을 15장 Presenter Slides로, 시각은 미니멀 등각 다이어그램으로 가자"

→ Skip the interview and deliver immediately. Only the cover and section divider slides are cinematic; the rest are diagrams.

Example 3 — Image prompts only

사용자: "S2 슬라이드 7번(TRUST-5 다이어그램) 이미지 프롬프트만 다시"

→ Regenerate only the corresponding slide block from Part B.

Checklist (self-verification just before delivery)

• Are all four NotebookLM axes (Format/Length/Language/Prompt) specified?
• Are all six Prompt blocks (audience, top-priority message, structure, tone, emphasized slides, prohibitions) present?
• Is the top-priority message one sentence? (If it is two or more sentences, the split failed)
• Are all 5 Components (Style/Subject/Setting/Action/Composition) filled in for every slide?
• Do all slide images share the same series tag(palette, lighting, style)?
• Is the aspect ratio unified at 16:9? (Default for lecture slides)
• Are real people's faces, copyrighted characters, or brand logos directly depicted?
• Are clichés ("innovative", "next-generation", "redefining", "in conclusion") absent from the NotebookLM Prompt?
• Is the AI slop post-processing (ai-slop-reviewer) scheduled to run?

Chaining recommended

media-notebooklm-slide-prompt
  ↓
moai-coworker:ai-slop-reviewer

Since this skill produces text output, right before the chain ends, moai-coworker:ai-slop-reviewerpost-process with.

Actual image generation is separated into a distinct step. The user executes the produced 5-Component prompt with one of the following tools:

• Gemini app (mobile·web)
• Google AI Studio
• Vertex AI
• (MoAI users) moai-media:media-higgsfield-image pass directly to the skill

Anti-patterns (what not to do)

• ❌ Pasting the body text wholesale into the NotebookLM Prompt field — the body must be uploaded as a notebook source. The Prompt field should contain only meta-instructions on "how to make it."
• ❌ Cramming two or more core messages into one slide — 1 slide 1 message principle.
• ❌ The Style of each slide's image prompt differs — series consistency breaks.
• ❌ Leaving Composition empty — the model's arbitrary cropping result doesn't match the slide aspect ratio.
• ❌ Directly depicting real people, real brands, or copyrighted characters — substitute with fictional characters, silhouettes, or metaphors.
• ❌ Instructing to reduce text when it's a Detailed Deck — the format and length instructions contradict each other.

Reference materials (Progressive Disclosure)

Instead of containing all guidance in the body, this skill separates the visual style library into a separate reference fileand loads it only when needed.

• references/slide-style-library.md — A library of 49 visual styles × 8 categories. Referenced during the tone decision step of Phase 1 and when writing the 5-Component Style fields in Phase 3. Built-in presentation keyword → style auto-matching rules, series consistency guards (Style·Palette·Lighting·Consistency tag), and anti-pattern catalog.

Style selection workflow:

[Phase 1 인터뷰] 사용자 발화에서 키워드 추출
        ↓
[references/slide-style-library.md "스타일 매칭 가이드"] 카테고리 매핑
        ↓
[AskUserQuestion] 카테고리 내 1~3 스타일 후보 제시 (영문 프롬프트 + 추천 상황 동봉)
        ↓
[Phase 3] 선택된 스타일의 영문 프롬프트 키워드를 모든 슬라이드 `Style:` 필드에 동일 적용

Official sources

• NotebookLM Slide Deck official documentation — Google Support
  https://support.google.com/notebooklm/answer/16757456
• Nano Banana (Gemini Image) prompt writing guide — Google DeepMind
  https://deepmind.google/models/gemini-image/prompt-guide/
• Nano Banana Pro / Gemini 3 Pro Image announcement — Google Blog
  https://blog.google/innovation-and-ai/products/higgsfield-image-pro/
• Nano Banana image generation — Google AI for Developers
  https://ai.google.dev/gemini-api/docs/image-generation

Note

• The original prompt builder design started from Goos Kim's "Harness Lecture Project" (2026-05) internal materials and has been reorganized.
• The 49 visual style classifications (references/slide-style-library.md) are reorganized categories, matching rules, and consistency guards referencing the publicly available Obsidian publish material "Notebook LM Slide Guidebook" (E-commerce Class, datawave).