Big Idea

The argument in one line.

NotebookLM's analysis of top-performing reference videos is the actual engine: it extracts the hook structure, sentence rhythm, and tone the algorithm already rewarded, then reproduces it on any topic at unlimited scale.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You want to start a faceless YouTube channel but do not know how to write scripts that match what already works in your niche.
You have tried AI image tools but your characters look different in every scene — consistency has been the blocker.
You want a repeatable production system you can run entirely on free tools without paying for Midjourney, ElevenLabs, or similar.
You are interested in the history/documentary/animation niche and want to see how the Lost Legacy model can be replicated.

SKIP IF…

You need high-fidelity character animation or motion beyond what Google Flow free tier outputs — this is a starter system, not a studio pipeline.
You already have a production workflow and are not looking for a from-scratch tutorial.

TL;DR

The full version, fast.

NotebookLM becomes a script factory when you feed it the top-performing videos from a reference channel and tell it to study their style. The key step is the analysis paragraph it generates before giving you ideas — that paragraph is what makes the output sound like the channel, not generic AI content. Pair the script with a visual planner prompt in a second NotebookLM project to get a complete scene table, then use Google Flow free daily credits to generate consistent character images and animations. The character name embedded inside every image prompt is what keeps your character looking the same across every scene.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:47

01 · Hook — Lost Legacy benchmark

Opens with the Lost Legacy channel as a proof-of-concept: 36 videos, 24M views, ~$80K from ads. States the promise.

00:47 – 03:53

02 · Step 1 — Script creation in NotebookLM

Load top 5 popular videos from a reference channel as YouTube sources, apply the Title and Script Creator role prompt, request 5 ideas, add web sources for the chosen topic, generate a 500-word cited script.

03:53 – 04:13

03 · Mid-roll CTA — Faceless YouTube Engine guide

First promotional beat for the creator paid guide. 9-stage pipeline, checklists, sample video built alongside reader.

04:13 – 06:13

04 · Step 2 — Visual planning in NotebookLM

Open a new NotebookLM project with the 2D Animation Guide PDF as source, apply the Visual Planner prompt, generate character image prompts and the full scene table.

06:13 – 08:27

05 · Step 3 — Generate images and video clips in Google Flow

Generate character images (naming each file after the character), build scenes by attaching character files, switch to video mode for animations. Explains the name-in-prompt consistency trick.

08:27 – 08:52

06 · Second mid-roll CTA — Faceless YouTube Engine guide

Second promotional beat for the guide. Targets people already making videos who want to cut production time.

08:52 – 10:04

07 · Step 4 — Assembly in CapCut

Voiceover goes in first as the skeleton. Scene clips placed per the visual plan table. Transitions, music from YouTube Studio Audio Library, export.

10:04 – 11:09

08 · Result demo + closing CTA

The finished 2D animation video plays. Closes with a comment CTA and next-video redirect.

Atomic Insights

Lines worth screenshotting.

NotebookLM generates an analysis paragraph before your video ideas — that paragraph is the engine. It studies what the algorithm already rewarded and injects that pattern into everything it writes.
Embedding a character exact name inside every image prompt in Google Flow is what keeps their face consistent across every scene. Skip the name, lose the consistency.
Google Flow gives you 5 free video generations per day per Google account. A second Google account doubles your daily quota — credits reset every 24 hours.
The visual planner scene table caps each voiceover line at 20-30 words. That constraint is structural: shorter lines force scene-sized chunks that match animation clip length.
Adding real sourced web articles before writing the script means every claim has a citation number. That is not just for accuracy — it is what keeps the channel safe from demonetization.
The entire tool stack (NotebookLM, Google Flow, Google AI Studio, YouTube Studio Audio Library, CapCut) is free. The only investment is time and daily credit limits.
The Lost Legacy channel model — harsh historical truths told from a second-person you-are-there perspective — earned around $80,000 from ads alone across 36 videos. The format is replicable.
Chunking the script into sections when generating TTS in Google AI Studio keeps the voice output stable and consistent. Long scripts in a single pass degrade quality.
For scenes shorter than the voiceover, duplicate-and-reverse is faster than re-generating a longer clip. The looped motion often reads as intentional animation style.
Swapping the reference channel and the topic is the only thing you change to start a new niche. The prompts, the structure, the workflow — all remain identical.

Takeaway

One prompt chain turns any top channel into your style guide.

WHAT TO LEARN

The insight is not that AI can write scripts — it is that loading the algorithm own winners as reference material makes the output calibrated instead of generic.

Sorting a reference channel videos by Popular before loading them as sources gives NotebookLM a signal-filtered dataset — you are training on what the algorithm rewarded, not what the creator uploaded first.
NotebookLM analysis paragraph, generated before the idea list, is the actual product. It extracts the structural and tonal patterns from the reference videos — that extraction is what separates this from a generic AI prompt.
Embedding a character exact name inside every image and animation prompt is the single step that prevents character drift across scenes. The name acts as an anchor the model can recognize across generations.
Google Flow free tier resets daily, making the credit limit a scheduling constraint rather than a cost barrier. Two accounts means ten video generations per day at zero cost.
The visual planner 20-30 word voiceover cap per scene is a production constraint that also solves an editing problem: clips generated in Flow are short, and the word-capped lines match their natural length.
Adding real web-sourced articles before generating the script creates inline citation numbers that serve as a factual paper trail if a video is ever flagged — accuracy and demonetization safety in one step.
The entire pipeline is format-agnostic: swap the reference channel and topic, keep every prompt and tool. The same system that produces roman gladiator videos produces cooking history videos without any structural changes.

Glossary

Terms worth knowing.

Google Flow: Google AI creative tool for generating images and short video clips from text prompts. Offers free daily credits for up to five video generations per account.
NotebookLM: Google AI notebook that lets you load documents, YouTube videos, and web sources, then chat with them. Used here as a style-aware script generator and visual planner.
Role prompt: A custom instruction pasted into NotebookLM Settings > Chat > Custom that tells the AI what persona to adopt and what rules to follow.
Visual plan: A three-column production table (Voiceover | Image Prompt | Animation Prompt) generated by NotebookLM, one row per scene, that serves as the assembly map for the final edit.
Demonetization-safe: Content structured to avoid YouTube automated demonetization triggers — no reused footage, no slideshows, unique visuals per video, original AI-generated assets.

Resources

Things they pointed at.

00:03channelLost Legacy (YouTube channel) ↗

00:56toolNotebookLM ↗

04:57productAI Garage 2D Animation Visual Guide (PDF)

04:10productThe Faceless YouTube Engine (guide)

06:14toolGoogle Flow ↗

08:59toolGoogle AI Studio (TTS) ↗

09:21toolYouTube Studio Audio Library ↗

09:34toolCapCut ↗

Quotables

Lines you could clip.

03:18

“Before the five ideas, notice this analysis paragraph notebook LM generated first. This is it studying the videos we uploaded, breaking down the style, the structure, the tone. That analysis is what makes everything that follows actually useful.”

Explains the core mechanism — not just what to do but why it works. Standalone and teachable.→ TikTok hook↗ Tweet quote

05:33

“Notebook LM picked that rhythm straight from the lost legacy videos we loaded as sources.”

Concrete proof point — shows the output is calibrated, not generic.→ newsletter pull-quote↗ Tweet quote

12:23

“That consistency comes directly from having the character name inside the image prompt, exactly what we set up in the visual planner step.”

Practical tip with a clear mechanism — highly shareable for AI creators.→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogystory

Hey. I want you to check out this channel, Lost Legacy. With only 36 videos, it has attracted nearly 24,000,000 views.

That is equal to around $80,000 from YouTube ads alone. And I'm sure the channel is also earning from sponsors.

So it's clear that this channel is earning pretty good money from two d animation videos. The good news is you can also create videos like this very easily. In this video, I'll show you how to build your two d animation video engine step by step.

We'll start by creating titles and scripts based on top performing videos. Then we'll turn the script into a full visual plan with scenes. After that, we'll generate the visuals and assemble the final video.

At the end, I'll show you the result. Let's start. If you're into breakdowns for video creation like this, hit like and subscribe.

I drop new videos every week. First, let's open notebook l m and create a brand new project. Now open YouTube.

We're going to use the lost legacy channel as our style reference, the same channel I showed you at the beginning of this video. Go to their videos tab and change the sort filter from latest to popular.

You're looking at the videos the algorithm rewarded most. These are your training material. Copy the URL of the first video.

Head back to NotebookLM and paste it in as a YouTube source. Repeat this for the next four videos so you have the top five URLs loaded in as sources. Hit next and let NotebookLM load them.

Now here's where the magic happens. Go to settings, then chat, then switch from default to custom.

Paste in the title and script creator role prompt. It's on screen right now, and it's also in the description. Save it.

Go to the notebook l m chat and type, give me five ideas for my next two d animation video. Before the five ideas, notice this analysis paragraph notebook l m generated first. This is it studying the videos we uploaded, breaking down the style, the structure, the tone.

That analysis is what makes everything that follows actually useful. Now let's go through the five ideas. I'm going with the first idea, daily life of Roman gladiators.

But before we write the script, I want to add reliable sources on this specific topic. On the left side panel, go to the search section and search Roman gladiator's daily life and hit the arrow. Let it search the web for this topic.

Then hit import to add the results directly as sources. Now when NotebookLM writes the script, it's drawing from real sourced material on this exact topic, not just making things up.

Back to the chat. Type. Write a 500 word YouTube script for idea number and put your chosen topic.

The script comes back fast. Scroll through it slowly. See those citation numbers sitting next to the claims?

Each one points back to the sources we just added. Every fact is traceable. That's what keeps this content reliable and your channel safe.

Let me read you the hook. Do you have what it takes to survive the coliseum? You probably picture a glorious champion.

You imagine gleaming armor and roaring crowds. But the truth is much darker. You are not a hero today.

You are a piece of property. This is a strong hook. Notice those short punchy sentences.

Notebook l m picked that rhythm straight from the lost legacy videos we loaded as sources. And here's the thing, this entire step is repeatable without limits. Swap the example channel, swap the topic, run it again.

Unlimited original ideas and scripts, all sourced, all styled to what actually works in your niche. Before step two, one quick thing.

I put together a guide called the faceless YouTube engine. Nine stages. No fluff.

Every stage has a checklist and a homework task so you keep moving. A real sample video is built with you throughout so nothing stays abstract. If you have never published a video before, this is where you start.

Link is in the description. Now you have a script. What you don't have yet is a visual plan, a scene by scene map of exactly what goes on screen.

Open a new Notebook LM project. Now upload the AI garage two d animation guide as the only source. This guide contains example scenes for the overall look and feel.

Download it as a PDF. The link to the guide is in the description. Upload that PDF into Notebook LM and hit the blue arrow to let it load.

Go to settings, chat, custom, and paste in the visual planner role prompt. On screen now in the description too.

This prompt turns notebook l m into a scene planner that reads the guide and outputs a production table with three columns. Voice over, image prompt, and animation prompt. Each voice over line is capped at 20 to 30 words per scene.

Before the table, it also generates image prompts for characters in your script. Quick reminder, this step works for any style. The only thing you need to change is the visual references guide.

The rest is exactly the same. Now go to the chat and type create visual plan for the following script. Then copy the script from the previous notebook l m and paste it.

Hit send. Now let's look at what came back. Before the scene table, notice these character image prompts.

Look closely. Every single one includes the character's name directly inside the prompt. That is not random.

In the next step, when we generate images in Google flow, that name is what keeps your character looking consistent across every scene. Same face, same style, every time.

I'll show you exactly how that works in a moment. Then the full scene table follows. Voice over, image prompt, animation prompt, row by row.

That's your complete production map. Time to bring the visual plan to life. Open Google Flow and start a new project.

Make sure image generation mode is selected. Let's start with characters. Go back to your visual planner notebook and copy the first character image prompt.

Paste it into Google Flow and hit generate. Once it comes back, rename the image with the character's exact name from the script. This keeps everything organized for what comes next.

Now the second character, same process. Paste the prompt, generate, rename.

I'll repeat this for all characters in the script. Here's a look at all of them created and ready to go. Now let's build the scenes.

Copy the image prompt for scene one from the visual planner notebook. Paste it into flow, then click the plus button and attach the character that appears in that scene. Generate.

Now switch flow to video generation mode. Copy the video prompt for scene one. Paste it in, and this time select the scene image you just created.

Hit generate. And there's the animation. Clean movement, exactly what the prompt called for.

Scene two. Same flow. Image prompt.

Attach the character. Generate the still. Then switch to video.

Paste the motion prompt. Select the scene. Generate.

Look at the character here. Same face. Same lines.

Same style as scene one. That consistency comes directly from having the character name inside the image prompt, exactly what we set up in the visual planner step.

Now a scene with multiple characters. Same process, but this time when I hit the plus button, I attach both characters to the scene.

Then I create the scene again. Look at both characters together. The style holds across all of them.

Here's the scene grid for the example scenes I generated. Before we move to the final step, I'm downloading all of them now. One thing worth knowing, Google Flow gives you free daily credits for up to five video generations.

If you need more in a day, use a second Google account for another set of free credits. Or simply wait. Credits reset the next day.

One more thing before I show you the final edit and the video. The faceless YouTube engine covers this entire workflow from choosing your niche all the way through publishing consistently.

Every chapter closes with homework, and a sample video is built alongside you at every stage, so nothing stays theoretical. If you are already making videos and want to cut your production time down significantly, the full system is in that guide.

Link is in the description. Script done. Visuals done.

Now let's put it all together. First, voice over. Open Google AI Studio and go through the voice options.

Since this is a story video, I want something with a lower pitch. Copy a chunk of the script, paste it in, and hit run. Download it.

Working in chunks keeps the output stable and consistent. I'll do the same for the rest of the script. Now background music.

Open YouTube Studio and go to the audio library. It's right there on the left panel. I filter by genre, cinematic, mood, inspirational.

Preview a few, download the one that fits. Now CapCut. Drop the voice over in first.

That's the skeleton. Upload all the scene clips. I pull up the visual plan and use it as my map.

It tells me exactly which scene goes where, so this part moves fast. For scenes shorter than the voiceover, I either duplicate and reverse or slow them down slightly. Add subtle transitions between scenes, drop the music in and lower the volume so the voice over stays in front.

Final check then export. Let's see the result.

Do you have what it takes to survive the coliseum? You probably picture a glorious champion. You imagine gleaming armor and roaring crowds, but the truth is much darker.

You are not a hero today. You are a piece of property. To keep your head attached to your neck, you must master the brutal reality of the Roman arena.

Welcome to the lowest tier of ancient society. You belong to a disgraced class called the infamous. You cannot vote or leave a will.

You are completely owned by a manager.

You swear a terrifying sacred oath today. If you want me to build this step by step guide for a different niche, drop continue plus your niche in the comments. I'll pick the most requested ones for the next episodes in this series.

And if you wanna see this same approach applied to a completely different niche, click the video on your screen right now. It's one of the most popular videos on the channel.

I'll see you there.

The Hook

The bait, then the rug-pull.

Thirty-six videos. Twenty-four million views. Around $80,000 from ads alone. That is the Lost Legacy benchmark — a faceless 2D animation channel built on historical storytelling — and the premise of this tutorial is that the exact engine behind it is now copyable, for free, by anyone with a Google account.

Frameworks

Named ideas worth stealing.

00:32list

3-Step 2D Animation Engine

Step 1: Title and script creation in NotebookLM (reference channel as style source + web sources for accuracy)
Step 2: Visual planning in NotebookLM (character prompts + scene table with voiceover/image/animation columns)
Step 3: Generate assets in Google Flow + assemble in CapCut

The core repeatable workflow. Change only the reference channel and topic; the prompt structure and tool stack stay constant.

Steal forAny faceless YouTube channel concept — swap niche, keep the system

02:23concept

Title and Script Creator Role Prompt

Custom NotebookLM system prompt defining: hook formulas, sentence rhythm (short punchy 6-12 word sentences), tone (upbeat, confident, conversational), structure (hook > problem > solution > takeaway), endings (clear takeaway or forward-looking statement).

Steal forAny AI tool that accepts a system prompt — reusable across ChatGPT, Claude, etc.

04:57concept

Visual Planner Role Prompt

Instructs NotebookLM to output Block 1 (character image prompts with name embedded) and Block 2 (scene table: voiceover 20-30 words max, image prompt, animation prompt).

Steal forAny scene-by-scene production planning for animated or illustrated content

07:07concept

Character Name Consistency Trick

Embedding the character exact name inside every image and animation prompt in Google Flow keeps face and style consistent across all scenes. Without the name anchor, style drifts between generations.

Steal forAny multi-scene AI image generation workflow with recurring characters

CTA Breakdown