Transcription
Hello and welcome back!
Did you know that Notebook LM can generate perfectly natural-sounding audio of two people discussing any topic you like with just a click of a button? The voices are AI, of course, and the information that they're discussing is based on any source you provide, such as a web page, a YouTube video, a Google Doc, or even just pasted text. The output is amazing!
Now you can use this audio to create podcast episodes and even video podcast episodes by using realistic-looking video avatars. So I'm going to show you how to do that in this video. It's really fascinating stuff and well worth checking out.
So let's start off by showing you an example of a video podcast made with two video avatars. Before we dive into the specifics, let's just define what an AI avatar is because maybe not everyone knows. Um, so it's basically a digital version of yourself, right? Like a digital twin that can be in videos.
Exactly! It's so cool. Yeah, so this tutorial walks us through how to actually create your AI avatar. Oh, and it seems really simple! You can choose a still image or an animated one.
Oh, so you could use a video you've already recorded?
Yeah, exactly! Interesting. So let's look at how to make a podcast in Notebook.
So we just go here to Notebook. The link is in the description as always. You create a new note, then you add your source. We're going to do it from a YouTube video, and we're going to use this video of mine about how to make a personal video avatar with Hey Jen. The link is in the description and in the card that will appear now.
So you just copy the link from YouTube like so, then click YouTube here, paste the link, and then you can click this big blue magic button to generate the conversation. In a few minutes, we have that ready, and we can play it back.
Okay, so have you ever thought about having like a digital twin making videos for you in different languages even?
Yeah, well, that's what Haun says they can do, and we're going to deep dive into it today.
It's pretty futuristic, it does, right?
Yeah, we've got this YouTube tutorial we're looking at, and they're saying it only takes 10 minutes to get started.
10 minutes, huh?
Yeah, so we're going to unpack like what it is.
Wow, that's just so cool! It's insane! What I love about it is the conversational style. I mean, the subject of the video is presented in a totally different way to the original video itself, which is a demonstration video.
So it brings the concepts to the viewer or listener in a completely different packaging—just a relaxed conversation, very human and relatable.
Now imagine how quickly you could produce realistic-sounding podcast episodes and start building up a following. In a few minutes, you've got an audio podcast that's ready to go and publish on the many popular podcasting platforms out there, like for example, Apple and Spotify, which are totally free and have millions of monthly listeners.
And you'll have no shortage of source content because you can take news articles, videos, stuff from anywhere on the web, and create an interesting dialogue about the subject. I've just started off, and here's my first episode now published on Spotify.
Okay, so that's the process for making the audio part of the podcast, and you could just stop there. Plenty of podcasts are audio-only. But if you want to take it to the next level and make it into a video, then you can do that with AI to turn it into a Talking Heads video.
Now, there are two ways I'm going to look at doing that in this video because I've tried them both. But I know there are other tools out there. However, if anybody knows of a free way to do this, then please leave a comment below, and I'll look into it.
So the first method is by using avatars made in Haen, and the second way is by using Hedra to make character videos. There are timestamps in the description, so you can jump to Hedra if you want.
For now, let's get stuck into the Haen method. First off, I just want to show you a bit of this podcast that I already made with Haen, and then we'll run through the steps I took to make it. You'll see that it's slightly different from the one I showed you at the beginning of this video, and the reason for that is that I used two different techniques within Haen to generate the avatar.
I used one avatar that was a photo avatar generated with a prompt, and I used another that was from their avatar library. Anyhow, here it goes.
But there's this one feature that totally blew me away.
Oh, tell me!
It can translate your script and your voiceover, wait, really, into 40 different languages!
That's insane!
I know! Think about what people could do with that, right? You could reach a global audience, so I think that is really cool.
Okay, so let's retrace my steps. First off, the podcast was created from this video. All I had to do was provide the video link to Notebook, as we saw earlier.
Right now, once you have the audio generated, you need to download it by clicking here, and then you need to bring it into Haen.
Now, with Haen, you can obviously make your own personal avatar. After all, that's what the example tutorial video is about. And you could, of course, choose to use that for your podcast video.
However, this process is more time-consuming because you'll want to use your own cloned voice rather than the voice that comes out of Notebook LM. So there are more steps involved. That's why I didn't do that with my example podcast video.
To keep this video as concise as possible, I'll show you how to do it with the Haen avatars. And if you want me to do a video on making it with your own avatar, then write me a comment below.
So you want to choose your avatars. You'll need a girl and a guy, and you can do this a few ways. You can pick from the available avatars, or you can generate a new photo avatar.
As I mentioned in my video, I did both. Mumi here is a photo avatar, while Vernon is from the avatar library. I do like the avatars that are available in the library because they have different angles with the same look—something you can't get when you generate your own photo avatar.
Also, if you don't have a perfect lighting setup at home or in your studio, a readymade avatar could be the best choice because it's so well lit with a nice interior design, etc.
So the basic process here is to make an avatar video using the podcast audio recording, but we need to make two separate videos, one for each avatar.
With the entire recording, then we need to bring them both into a video editor like CapCut and do a bit of manual chopping and cutting. An alternative way of doing this would be to chop up the audio first to get an audio track for each avatar, then import those tracks into Haen to make each video.
I'd use Audacity for that, as it's free and open source. Link in the description. There is an advantage to separating the audio first, and that is that you could have both avatars on screen at the same time.
However, as I did my podcast video cutting from one avatar to the other, I'll show you how I did it that way. Then when we look at the Hedra method, I'll show you how to separate the audio first.
So let's get through this.
Okay, so inside Haen, the first step is to choose your avatar. And as I said, we want one boy, one girl. I chose Vernon, who's one of these video avatars from the library. Here he is because he has this nice look where he's wearing this nice suit, and he's in a great interior.
I really like this interior, but not only that, there's a second angle of the same look, so that meant I could cut between two angles in the finished video.
Then once you've chosen an avatar from here, you just click this button here, create with AI Studio, choose landscape if you're doing a landscape video, and it will load the studio.
So from here is where I would create the video, but let's first of all go back to the other avatar before we do that. The other avatar is a photo avatar, which I created because I wanted to have an original avatar that wasn't available in the library.
So I created Mumi here and all these different looks, and in the end, I chose this one to use. So again, from here, I can click here and go straight into the editor.
All right, so let's do that, and I'll show you how to make the video.
So we have loaded. All we need to do is come down to script, and we have a voice and a pre-loaded script here ready. But we don't want those because we need to use the audio from Notebook LM.
So we'll just delete this script and then upload. You just click here, and it will open up a browser window for you to select the audio where you have it saved on your system.
Okay, so I'm just going to select this, and so we have the audio loaded here. And then you can preview the audio.
Okay, so have you ever thought about having like a digital twin?
Okay, so now what will happen is when we submit this video to be processed, Haen will lip sync Mumi to this audio.
Now, just one thing to note: the original audio from Notebook LM is just over 10 minutes long. Now, depending on the plan that you have with Haen, you may have a time limit.
Okay, so I'm on Creator at the moment, and I have a time limit of 5 minutes. If you're on free, it's a 3-minute limit. If you're on team, then it's up to 30 minutes.
So if I was on team, I could process the whole video in one go. But as I'm on Creator, I've had to do it in three parts because it's just over 10 minutes.
Okay, so that's just something to take into account. It's a few extra steps if you don't want to spend the extra money.
Okay, so now that's ready. Literally, all we did was load Mumi, delete the default screen, and upload the audio. And that's it!
Then we click submit, we give it a name, 25 frames per second, resolution 1080p, format MP4, and the Haen watermark is off. And then we submit.
The process for the video avatar is exactly the same. You go inside here, you select the look that you want, click create with AI Studio, choose landscape.
Here we do have a proper 16x9, so I don't have to do any extra editing. Go to the script, delete the default, upload the audio, and that's literally it!
I can submit, and I'll have the Vernon version of the video too. Just to note, in both videos, we have both people talking to it.
Today, it's pretty futuristic, it does, right?
Yeah, we've got this YouTube tutorial we're looking at, and they're saying it only takes 10 minutes to get started.
10 minutes, huh?
Yeah, okay, as we said, this method involves making a video of each avatar with the audio of both speakers and then splitting them afterwards rather than doing it beforehand.
Okay, and once you hit submit, the video normally takes a very short space of time, like a few minutes, to process.
So just to show you the finished videos, here's Vernon and the second part tutorial.
Let's do it!
All right, let's go!
Okay, so we've gotten a taste of what Haen can do.
Yeah, it's pretty impressive!
Okay, and here's the version of him from the other angle.
Well, I think it's important to remember that AI is just a tool.
Okay, it's how we use it that matters.
And the videos of Mumi, well, I think it's important that matters, right? Like it's not about replacing us exactly; it's about augmenting what we can do.
Okay, and then the last step in Haen is just to download the videos. So you go up here to three buttons, click download, and then download it to wherever you need to save it on your system.
Right now we have both videos, we can bring them into a video editor. I'm going to use CapCut, which is a really powerful editor with a huge amount included in the free forever plan—more than enough to do what we're going to do. The link is in the description.
So once you have the clips imported, you just layer them on the timeline so they're playing simultaneously.
Have you ever thought about having like a digital twin?
Then you make a cut where the speaker changes from one to the other.
M in different languages even?
Yeah, and you delete the part of the track that you don't want. You alternate like this from one track to the other until you have the whole video done.
And here's what the timeline looks like when finished. Note that I have two tracks for Vernon due to his two angles, which made this video a bit more complicated, but it's not necessary to do that.
Anyway, let's just take a quick look at it.
I found it super interesting!
Oh, what's that?
It's called the consent recording process.
H consent recording? That's new!
Yeah, so you...
Okay, so there you have it. This is the most fiddly part of this whole process, but it's not too bad once you get into it.
It's still an incredibly fast way to go from having a source to a finished podcast video.
Now let's take a look at the Hedra method.
So for the Hedra method, we'll be making some characters with Hedra to use as our Talking Heads. But remember, this time we'll separate the Notebook LM audio into two separate tracks, one for each speaker.
That way we can have both speakers on screen at the same time.
So to split the audio, I imported it into Audacity, free and open source audio software. Link in the description.
Then using the keyboard shortcut Command I or Control I on Windows, you split the audio every point where the speaker changes from one to the other.
Have you ever thought about having like a digital twin making videos for you in different languages?
Even different languages, even?
Yeah, and then you just use Command D or Control D to duplicate the track, label one as Mumi and the other as Vernon.
From there, you simply delete the surplus parts so that Mumi's track only has Mumi's voice and Vernon only has Vernon.
Your timelines should look like this when you finish.
Once you have this done, you can export the audio and then go to Hedra. Link in the description for Hedra.
Here, the first step is to describe your first character, as I've done here: a photo of a friendly woman wearing headphones in a recording booth, body and head both facing directly towards the camera.
Then hit create and see what it generates for you.
I like this output, so it's a keeper.
Now all we need to do is to upload the audio that we just prepared. We do that over here in the audio section by clicking this upload audio button.
Okay, so have you ever thought?
Then just hit generate video, and here's the generated video. We'll have a quick look at that.
Okay, so have you ever thought about having like a digital twin making videos for you in different languages even?
Well, that's what Haun says they can do, and we're going to deep dive into it today.
Okay, fine!
And now we'll do the same for Vernon.
Now we have Vernon done as well, right?
'Cause everyone's talking about AI these days.
Yeah, so we need to cut through that.
And lastly, we just need to bring both clips into any video editor. Again, I'm going to use CapCut, and all you need to do is to place both clips on the timeline and separate them so that one's on the left and the other's on the right.
If you press play, you'll see the effect of both of them working together.
In different languages even?
Yeah, well, that's what Haun says they can do, and we're going to deep dive into it today.
It's pretty futuristic, it does, right?
Yeah, we've got this YouTube tutorial we're looking at, and they're saying it only takes 10 minutes to get started.
10 minutes, huh?
Yeah, so we're going to unpack like what it is and how it works.
And I think the effect is really nice!
So we've looked at how to generate an audio dialogue from any information source using Notebook LM, and we've also looked at how to turn that into a video using realistic video avatars.
Now, can you think of another application for this audio? If so, please let me know in the comments below. I would love to hear about it.
And if you found this video useful, please consider subscribing.
Thanks for watching!