I Turned NotebookLM Into a History Channel Engine (Full Workflow)

Tech + workflow summary (NotebookLM → “history channel engine”)

The video presents a full, step-by-step automation workflow to produce myth-busting history videos using AI tools. The goal is to generate an entire script, visual plan, images/video clips, voiceover, and the final edit with minimal manual work.

Core idea

Proposition

You can create a working “history channel engine” starting with NotebookLM by:

1. Training it on sources (successful history channels)
2. Configuring it with a role to act as a fact-checker
3. Feeding it research articles
4. Generating:
   • a script
   • a visual plan
   • batch visual assets
5. Producing voiceover and assembling the final video

Step-by-step tutorial / build instructions

Step 1 — Build the “research engine” in NotebookLM

• Action: Create a new NotebookLM project.
• Key move often omitted: Feed NotebookLM real, successful long-form history channels first to learn their “rhythm and tone.”
• Sources loaded (3 channels):
  • Kings and Generals
  • The Armchair Historian
  • The Entire History
• NotebookLM settings: Add a custom role message so it behaves like a:
  • “fact-checker and myth-buster scriptwriter”
• Purpose: It should already understand the format before ideation.

Step 2 — Generate a history “theme” (myth topic selection)

• Ask NotebookLM for 5 debunking topic ideas, each including:
  • a myth and its correction (in one line).
• Example topics mentioned:
  • Cyrus Cylinder / human rights claim
  • Clumsy knight / medieval armor claim
  • Operation Long Leap / Tehran assassination plot claim
  • Golden Age pirates / cavalry obsolescence in WWI
• The example chooses: “cavalry obsolescence in the First World War.”
• Non-skippable step: Use NotebookLM in-depth research by importing 2–3 specific articles related to the selected topic.
• Auto output detail: It finds candidate articles (e.g., subtitles mention “about 20 fonts,” likely papers/sources) and imports the ones that fit.
• Goal: Make facts traceable and contrast “good-sounding” vs “correct” scripts.

Step 3 — Write the script with verification labels

• NotebookLM generates an ~800-word script.
• Script structure described:
  • Opens with a cinematic image (cavalry charging into machine-gun fire)
  • Argues cavalry wasn’t truly obsolete
  • Explains mounted units often fought as rapid reaction infantry (dismounting rather than saber charges)
  • Traces the myth to early WWI visible disasters (example: Battle of Halen)
• Verification labels are placed precisely on:
  • specific dates
  • units
  • casualty figures (Not on portions that are already established.)

Step 4 — Create a visual plan (shot list) from the script

• In NotebookLM settings, swap in a visual planner prompt.
• Generate a scene-by-scene plan within the same project.
• Output described:
  • “Eight locked characters and environments” (knights, warhorses, tanks, battlefield locations)
  • A table of 36 scenes, one row per scene, ready to export.

Step 5 — Batch-generate images and animated clips in Google Flow

• Tool: Google Flow (new project)
• Model: “Nano Banana 2” image model (selected in prompt settings)
• Aspect ratio choices:
  • 16:9 (long format)
  • 9:16 (shorts)
• Agent option: enabled (used to orchestrate batch generation)

Lock references / consistency

• Generate reference assets for characters/environments (e.g., knight, warhorse, German Uhlan, trench landscape, Halen battlefield).
• Use “insert scenes” under each reference so appearance stays consistent.
• Drag each reference image into the prompt box so Flow maintains the exact look across scenes.

Batch prompt workflow

• Insert all scenes using a full list of image prompts.
• Add instruction to rename each generated image by scene number.
• Switch to image → video mode:
  • set clip duration
  • generate animated scenes using the video prompt + full ordered image list to the agent
• Flow displays credit cost in advance.

Credit constraint noted

• Image generation: “unlimited”
• Video generation: limited by credits
  • Example: ~12 credits per 8-second clip
  • Example: 50 free credits every day

Step 6 — Convert the script to voice (ElevenLabs)

• Tool: ElevenLabs Text-to-Speech
• Model: V3
• Voice guidance: calm and clear (myth-correction tone, not horror/exaggeration)
• Fragment strategy:
  • Split the script into fragments under 100 words
  • Generate and listen to each fragment before moving on
  • Download fragments and number files to preserve order
• QA attention: check for mispronunciations and monotony.

Step 7 — Assemble the final video in CapCut

• Preparation: Put everything into one folder:
  • numbered voice clips
  • numbered video clips
  • music tracks
• CapCut import: import the whole folder at once.
• Editing order:
  1. Place voiceover first
  2. Remove silences
  3. Create a composite clip from selected voice clips (to prevent subtitle misalignment)
  4. Place animated video clips on the timeline to match narration lines
  5. Trim each clip to fit the corresponding line
• Transitions: choose scene-appropriate transitions for fluid cuts
• Subtitles:
  • CapCut subtitles → generate → choose a template (white text with black outline mentioned)
• Music volume: around -20 dB
• Finalize: preview the full video, then export.

Review / analysis angle

• The workflow is framed as an “engine” comparison against luck:
  • successful history niches succeed due to repeatable format
  • the process replicates effective storytelling cadence automatically
• Claimed efficiency win:
  • the full planning/research/script/shotlist pipeline runs inside one NotebookLM project
  • fewer tool jumps across those steps
• Practical constraints addressed:
  • Flow video generation credit limits and per-clip cost
  • chunked TTS segments + strict file ordering
  • a CapCut composite-clip step to avoid subtitle issues

Main speakers / sources (as mentioned)

• Speaker/source: the creator, “AI Entrepreneur” (host/voice of the tutorial)
• History-channel sources used as NotebookLM style inputs:
  • Kings and Generals
  • The Armchair Historian
  • The Entire History