Transcription
Most creators spend hours on a single video, researching, scripting, thinking about visuals. Hours, sometimes days, every single time.
What if that entire process ran itself? Today, I'll show you how to turn Notebook LM into a full video engine, configured in a way most people don't know is even possible. It researches your topic, writes your script, and feeds directly into a production pipeline that costs you exactly nothing. Free tools, unlimited output. And at the end, you'll see a sample video this system produces. Let's build it.
Open Notebook LM. Create a new notebook, but don't touch it yet. First, we need to discover what the internet actually wants to watch right now. Open YouTube in a new tab. Type one word in the search bar, psychology. Hit search. Now, go to filters. First, sort by popularity. Then, filter by upload date, this month only. You want what's winning right now, not 6 months ago, not last year, right now.
Look at the results. Scroll through them. At the very top, there's a video about the psychology of dog sleep. That topic is pulling serious views inside the psychology niche this month. That's your signal. Now, go back to the search bar. Type psychology of dog sleep. Apply the exact same filters, popularity, this month. You're now looking at the top five results inside a validated, actively trending sub niche. Let's grab these links one by one.
I'm right clicking on the first video, copy link. Going back to Notebook LM, pasting it in. Back to YouTube, right clicking the second video, copy link. Back to Notebook LM, pasting it in. I am doing the same for other videos. Five sources in, but we're not stopping there.
Now, open Google. Search the same phrase, psychology of dog sleep. You'll see a mix of articles, studies, and editorial breakdowns at the top. We want five of these as well. First link, copy, back to Notebook LM, paste. Second link, copy, paste. And doing the same for other links. After having all links, I hit generate. Now, your notebook has valid sources, top YouTube videos and top articles, all covering the same topic from different angles. That's not a guess. That's a knowledge base built from what's already winning.
Here's where most people stop. They have sources. They start chatting with the notebook. They get summaries. They get disappointed. The problem is that Notebook LM, in its default state, behaves like a research assistant. It answers questions, summarizes content, connects dots. Useful, but not what we need. We need it to write a show.
Click on the settings icon inside your notebook. Navigate to the chat section. You'll see an option labeled custom. Click it and paste in the script writer configuration prompt here. I'm showing it on screen now. You can also find it in the description and pinned comment. What this does is transformative. You're not changing the sources. What you're changing is how Notebook LM processes and communicates what it finds in them. With this prompt active, the notebook now has a writer's persona. It has a hook formula. It knows what pacing feels like for YouTube. It knows how to open with tension, how to sequence information to keep someone watching, and how to close with momentum. And critically, every claim it makes in the script still traces back to those 10 sources. This is what separates a notebook full of notes from a notebook that produces content.
Go to the chat tab. You don't need a long or complicated prompt here. The hard work was in the configuration. Type something this simple. Write a 1,000-word YouTube script about the psychology of dog sleep. That's it. Hit enter. Watch what it produces.
Let me read you just the first few lines of the hook it wrote. You probably think your dog sleeps in your bed because your mattress is soft. You are wrong. It was never about the thread count or the plush pillows. When your dog circles three times and curls up next to you, it is an ancient biological pact. Did you catch that? That's not a summary. That's a hook engineered to stop a scroll. Notice the specificity in the first line. It makes a claim that feels surprising. That's what makes someone lean in instead of clicking away.
Scroll through the rest of the script. You'll see the structure underneath it, an opening tension, a layered explanation, a key insight that reframes everything, supporting evidence, and a closing CTA that invites action. Every section earns its place. At the bottom, you'll find a summary of the most important insights from trusted sources you uploaded. So, with this system, you can be confident that every script is built on proven facts. This took less than 60 seconds to generate, and it's sourced. It didn't make anything up. It pulled everything from trending content we loaded in step one.
Copy the full script into a doc file. Open ChatGPT and paste the prompt you see on screen. You can also find it in the description and pinned comment. At the end of the prompt, you'll see a section labeled script. Simply copy your script and paste it there. This prompt instructs ChatGPT to behave like a video producer. It reads through your script line by line and assigns a visual type to each moment. What you get back is a production table, a scene-by-scene breakdown of your entire video. The table tells you exactly where to use each of the following.
Host on camera for moments where credibility or personal connection matters. These can be delivered by a real host, an animated avatar, or a static image host. Text overlay or graphic for key stats, definitions, or bold claims that land harder as words on screen than as spoken audio alone. Stock footage for environmental or emotional context. Sleeping dogs, nature, abstract scenes that support the mood of the narration. AI-generated visuals for anything conceptual, futuristic, or scientific where a stock clip doesn't quite fit. This table is your production blueprint. Every asset we build in the next section comes directly from this document. Have it open the entire time you're producing.
Now, we build. Let's go category by category using the table as our guide. For example, to create the first text overlay, open Canva and add text on a black background. Copy and paste the text. Choose a bold, YouTube-style font. Use a large size, high contrast, and a clean layout. Add a simple animation, then export it as a PNG. Done in 2 minutes.
Now, let's create the next scene using stock footage. Go to Pixabay and switch to photo search. Copy a keyword from the table and search for it on Pixabay. Browse the results until you find a suitable image, then download it for free. Let me show you an AI visual example. Open Google Flow and select image generation. Copy the description from the table and paste it in, then generate the image. Notice the quality and detail. Scenes like this grab viewers' attention and keep them engaged with the video. Download it for free. Repeat these simple processes for every scene.
For voiceover, go to Google AI Studio. Pick a voice that fits the tone of your content. For a psychology niche, something measured and clear works well. Now, here's the part most people get wrong. Don't paste the entire script at once. Break it into sections. If you dump the full script in one block, the quality degrades toward the end. The pacing gets lazy. The emphasis flattens out. Start with just the hook. Copy and paste it here. Generate.
Let's listen to it for a few seconds to test it. You probably think your dog sleeps in your bed because your mattress is soft. You are wrong. It was never about the thread count or the plush pillows. Hear the breathing between phrases? That's not robotic. That's a voice that sounds like someone who believes what they're saying. Download that clip. Move to the next section. Repeat until you have voiceover for the whole script in chunks.
Open your video editor. CapCut, Premiere, DaVinci, doesn't matter. Import everything. Your voiceover segments, your AI clips from Flow, your stock footage from Pixabay, your text graphics from Canva. Here's the workflow. Lay the audio down first. Lock the track. Now, use your production table from ChatGPT to drop visuals on top, exactly where they belong. The table already told you what goes where. You're not guessing. You're executing a plan. Underneath everything, add a low-volume music bed, atmospheric, minimal. Your voiceover should always be the loudest element in the mix. Finally, add a simple zoom effect to some scenes. Nothing fancy. A soft zoom makes it look more professional.
Let me show you a short clip from the assembled section. You probably think your dog sleeps in your bed because your mattress is soft. You are wrong. It was never about the thread count or the plush pillows. When your dog circles three times and curls up next to you, it is an ancient biological pact. Today, we are decoding the fascinating psychology of dog sleep. You will learn what their specific sleeping positions actually mean.
That is a real video. A complete, watchable, sourced piece of content built entirely with free tools. If you have questions about any step in this process, drop them in the comments below. I read everything and I'll reply directly. And if you want me to keep building these end-to-end system videos, just type continue in the comments. That one word tells me exactly what to make next.