Loading summary
Transcript99 lines
- [00:01]
Michael Stelzner
Looking to transform your marketing with AI? Social Media Marketing World 2025 has all the AI training you'll need. Join me and top AI experts in San Diego this March. Get your tickets now@socialmediamarketingworld.info and save big. Welcome to the AI Explored podcast, helping you put AI to work. And now, here's your Michael Stelzner. Hello, hello, hello. Thank you so much for joining me for the AI Explored podcast brought to you by Social Media Examiner. I'm your host, Michael Stelzner, and this is the podcast for marketers, creators and business owners who want to know how to use AI. Today I've got an exciting show for you. I'm going to be joined by Aoife Roche and we're going to talk about how to create AI avatars. These are video clones of yourself that look like you and sound like you so that you can use them to create all sorts of different content. If you're new to this podcast, be sure to follow this show so you don't miss any of our future content. Let's transition over to this week's interview with Aoife Roche, helping you simplify your AI journey. Here is this week's expert guide. Today, I'm very excited to be joined by Aoife Roche. If you don't know who Aoife is, she is a marketing consultant and AI strategist who helps business owners solve problems with AI. She's host of the Business Innovators podcast. She's also the founder of Super Dupe AI, an all in one video avatar content creation platform. And her membership is called the AI Authority Lab. Aoife, welcome to the show. How you doing today?
- [01:49]
Aoife Roche
I'm good, thank you. Thank you for having me. I really appreciate it.
- [01:53]
Michael Stelzner
I'm excited to talk to you. Today we're going to explore how to create AI video clones, which is really exciting. But before we go there, I want to hear your story. How'd you get into AI?
- [02:03]
Aoife Roche
Yeah, it's like because everyone became an AI expert overnight, right? Isn't that what happened with ChatGPT? But no, my background is in corporate, right? So large scale tech, transformational change. So I spent the best part of 10 years in the UK building up that career, being brought into large financial institutions to look and help them with integrations of technology across their industry wide and company wide. So then I left that and became an entrepreneur in 2018 because I realized marketing was very similar where data was just moving from left to right, Right. So we had to figure out how to move the data and when chatgpt came out and I started looking at the levels of automation and technology. I was like, okay, this is very familiar because, you know, the idea of the integrations of figuring out how we could optimize processes and procedures and do all of that. So then I just started really getting much more involved, knowing that with the way that the AI is developing and advancing, anyone who's a marketing agency owner right now really does need to shift to be a data driven marketing agency because AI pretty much executes on 80, 80 to 90% of the work that US agencies used to. So it was a strategic move as well as one that was a good fit.
- [03:23]
Michael Stelzner
Okay, so how soon after ChatGPT rolled out did you start realizing, maybe I can have my business be very focused on AI and kind of tell us about the journey and all these different things that you've, you've been creating.
- [03:37]
Aoife Roche
And I tell you, it started because we had a team of. At one point, at the beginning of 2023, I had 31 people. So it was quite a large operation with we had 26 virtual assistants running and then five people across. It was a lot. So when Chachi Boutique came out and I started looking at being able, process orientated and looking at what we could actually execute with the, you know, the, the advancement of AI, we scaled, we're down to. We have four virtual assistants now, down from 26. So. So that was the kind of the real threshold moment for me when I said, well, if I can do this for me, you know, and it's not that I want to be trying to get rid of jobs and get people out of their jobs, but this is about creating efficiency, right? How can we do things better, faster and more efficient? And then once, once I saw that in my own organization, then I realized the vast majority of people had the exact same problems. So then I started going out and becoming a much more prolific voice, I think on talking about how they needed to become data driven.
- [04:40]
Michael Stelzner
Well, and tell us about these different various things that you started along the way as well. Super dupe.
- [04:44]
Aoife Roche
Oh yeah. So well even, you know, at every layer and level now. So we had a copywriting team. We got rid of them. AI. ChatGPT is now our copywriter. We had an SEO team. We got rid of the SEO team. We now use Siona AI for our SEO. We had a graphics team. Now the graphics. I still don't think Canva. Canva's doing a good job, but they haven't exactly got there. We had an entire video editing team. Now we still have the video editing team. But with the combination of Opus Pro and Submagic, we've massively reduced sound. So it was like looking at every single vertical within the business business, then looking at every single task that was being executed within that and then looking for the opportunity for automation within that task. Not everything can be automated 100%, but that's what we did. And while I continue like every day, there's so many new AIs out there. Like I'm like, I saw one recently, was a 30 day AI challenge and I was like, I nearly do that, only I'm too afraid I'll go down another rabbit hole and I'll end up spending more and more time. But one of the things that kind of brought me to the super dupe was we had an agency that used to do med spa marketing and I had to do a lot of VSLs for the ads and I hated it.
- [06:00]
Michael Stelzner
Video sales letters, is that what you're talking about?
- [06:02]
Aoife Roche
Yeah, exactly, yes, sorry, Video sales letters, where you have. For the funnels. I had to do these videos and then I had to learn the script, hire the videographer, get them to come around, do my hair, do my makeup. And I felt under such pressure to do it and I hated it when it was recorded and then I'd make mistakes and I thought, come on now, there has to be a better way to do this. And that's kind of where the super dupe idea kind of was born out of. Was. Okay, well what if we could create a version of me that looked like me, talked like me, sounded like me, moved like me, but wasn't me, so that I could create these mass marketing video sales that are email campaigns or content for social media. And that's kind of how that idea grew. It grew from a desire to not want to be on social media.
- [06:49]
Michael Stelzner
Very fascinating. And then tell us about the AI authority lab also.
- [06:52]
Aoife Roche
Yeah, so with that. So I've been running a bunch of training programs for like eight weeks with people where I teach them how to leverage LinkedIn and AI to optimize their LinkedIn presence, to do their outreach and their lead generation. I've been running that. I put, you know, dozens of students through that. But now I've realized that it's just changing so rapidly. I was rerecording the course every six, eight weeks. Right. Because the AIs are changing so quickly. So then I thought, you know what, rather than putting people into these eight week courses, why don't I just put them into a 12 month membership program and then put all the courses on there. And as I create the courses, they'll get access to them under a paid membership. And that way I can have more flexibility to be able to put in more wider and broader content rather than just being LinkedIn. So that's definitely the newest initiative for 2025. And I'm excited for that because. Because it's going to allow me to really not have limitations on what I can teach people within a timeline or within a budget. It's like you join and you're all in for the year and you get everything I'm learning. You can learn as well.
- [07:57]
Michael Stelzner
Very cool. Okay, thank you so much for sharing your story. We're here now to talk really about making video clones of yourself. AI avatars, as they're referred to in the AI vernacular. The first question I want to ask you, which is a question I'm sure a lot of people are probably wondering, why in the world should I, as a marketer or a creator, clone myself in video? I know you hinted at it a little bit, but let's like unwrap that a little bit as to why that might be something people should consider.
- [08:26]
Aoife Roche
Yeah. So you know that public speaking is like one of the, you know, number. They say it's number one. Whether it is, we'll say top three. Right. Just for argument's sake of speaking in front of large groups of people. I reckon video marketing, doing video for your brand is probably a very close cousin of public speaking. Right. And most people do not want to do it. They just don't, unless they're actually a very specific content creator. And what happens then? We have a bunch of people who have great businesses, but they're just the best kept secret because nobody knows about them. And whether people like it or not, in the current economy, video marketing is the fastest way for you to be able to reach your customer. It's the fastest way you're going to be able to build authority, build brand awareness, educate people on what you're doing. Because you have Facebook, Instagram, TikTok X, you know, streaming YouTube. YouTube, like there's so many. And it's all leaning into the video. So that's the first thing is if you are not doing video, this is sort of a hack for you to do video until you get the confidence to be able to do video. That's the first kind of one. Because the objections I used to get from people running an agency, I don't have the time. I don't know what to say and I don't like how I Look on camera. Those were my three objections. So I built the product around the objections. So I said, well what if I solved that you only had to look good once on camera, that I would script what you actually had to say and once you had done your two minute video, you would never have to re record it again. If I solved all of those problems, would you buy my product and I got an over resounding yes from people. So that's one reason if you're not someone who's already doing video. Another reason is simply volume and scale. Right? We can create masses, amounts of content with the avatars right now and then put that out where you could be doing a reel every five minutes if you really wanted to, like depending on the level of effort that you put in. So being able to put it out there and get in the brand awareness to be able to do and curate the content without all of that narrative that goes around it and then the next one is just the cost. Like I have someone right now, they're doing 800 minutes which is just shy of 10 hours. Well maybe we'll say 12 hours. I don't know if he's going to use the whole we benchmark for 10 hours and he's recording it all course content with his avatar, how long will it take him to sit down and record 10 hours of course content perfectly. And he's just every day going and creating videos and videos. So the time saver, the money saving, the convenience and just the level that you could take it to with brand awareness across all of the platforms, it's.
- [11:20]
Michael Stelzner
Pretty kind of fascinating. And you know this concept of AI avatars or clones, today we're talking about it as a actual like pre scripted kind of thing. But very soon this could be a live experience, right? This could be like an interactive experience with a quote unquote virtual salesperson. Right? I mean that's coming, right?
- [11:46]
Aoife Roche
It's already there. It's already there. Right now you can build a private like LLM where it's not, I wouldn't say it's private LLM for, but it is to you. You can put in your own data into another LLM, connect the avatar and then the avatar is interactive so that avatar can be put on your website and then as long as it has the answer in the back it will actually go and it will pull the data and it will respond like it's an interactive experience.
- [12:15]
Michael Stelzner
That's pretty crazy. That's really cool. Okay, so let's start by defining what actually is an AI avatar. Slash, video clone. Like, let's just define it at its most basic level so people can wrap their heads around it. And then let's talk about like, actually some of the cool things that you've been doing with them and some of your clients have been doing with them.
- [12:37]
Aoife Roche
Yeah. So what we do very specifically, because you'll see there's a couple of technologies out there where you can use other people's visual. Right. And some, some of them can look a little bit robotic. It's getting a bit better. But you can use other people's visual. Our concept is not. I am creating a version of you with your voice, your nuances, your like, I, I do a long blink. Apparently I didn't know I ever did a long blink until I recorded an avatar and the avatar started blinking for these extended period of time. I was like, oh, how interesting. I never knew I did that. But it's a. We are creating an actual version of you that looks exactly like you. Well, it is exactly like you because how the technology works, it takes a two minute video and then you, you put that on and then the technology rolls forward, records every nuance and everything, and then it just rolls back again and it just keeps doing that over and over again. That's how the technology actually works in creating the avatar. So when people say, well, you know, that doesn't look exactly like me, I'm like, no, it is exactly like you. You just don't know that you look like that because you've never watched yourself, you know, or your nuances. So it's very, very interesting when people, when they see themselves sometimes it can be very. But it is getting so close. Up until 2024, we were still doing a lot of work on trying to making sure that all of the nuances were being captured. But I would arguably say at the end of 2024, we hit it to the point I don't even think you can distinguish right now. But like before, if you had a side by side, you would be kind of able to tell. But we're getting to a point now where it's actually unrecognizable.
- [14:23]
Michael Stelzner
So we already talked about the example of a course that you could create. So just so I can wrap my heads around this, let's say that you're a writer or let's say that you're a podcaster, right. You can record your voice or you can script a course on something that you're an expert in. And then essentially is the idea here that you feed the script to the AI Avatar, Is that the concept we're really talking about here?
- [14:50]
Aoife Roche
Yeah. So we have two options the way that we do and we're about to launch a third one. So we have two where you can upload your own script, so you can go and create it, you can upload your own own one and it'll verbatim recite back exactly what's on that script. We have a scripting engine in built on the platform where you can actually just give it, you know, a headline. You can choose demographic styles, you can choose Personas, you can choose marketing frameworks and it will generate you the script. And the next one that we're moving towards now is that, you know, with the GPT agents and assistants, people are creating these in the background where they're pre trained models or pre trained assistance on a particular data set that you would actually be able to connect the GPT assistant and it will pull the information from that data set.
- [15:38]
Michael Stelzner
Well, and also what about the audio podcasters out there that have like these libraries of audio podcasts, but they never recorded video?
- [15:45]
Aoife Roche
And I really see this is the benefit I say for the introverted podcaster, right? So we have so many people who only do audio, but only doing audio right now is just, it's just not enough. Right. Like you see if you look on Instagram right now, all of the really big podcasts not only have their own accounts where they're doing short form content on Instagram reels, or they also have multiple sort of clipper accounts, right? So people have detailed discord channels and then they go in and they upload the video and then these people clip them. And you could find, not to use Andrew Tate as an example, right? There was thousands of profiles out there pretending to be Andrew Tate and they're clipping his video content and they're circulating that out and that's how he's getting such the volume and the views. But if you're a podcaster and you're not doing the video, you're missing out on that opportunity that those little bite sized pieces of really great content are actually going to be heard by a wider audience or seen by a wider audience. Because not all of your audience is going to be auditory, some are going to be visual. And it doesn't matter how good the audio you give them, if they don't have that visual experience to be able to connect with you, they're not really going to subscribe into it. So with the idea of what you can do with the podcast is you could take the podcast, find the clips on it, like use AI to clip the, you know, the potential viral moments. Have your avatar created and then just load up that and it's going to automatically create the video for you. So then that video can go out on social media. So no podcaster ever has to sit there and say, I, you know, I'm not doing video because I don't want to do video. Because you only need to do the video once for two minutes and then it's there for the rest of of the time you actually want to put out that content.
- [17:29]
Michael Stelzner
Just so I understand, what we're not talking about is taking an audio only podcast and generating a video that is the voice of the actual podcaster, right? Or are we talking about, well, you could.
- [17:41]
Aoife Roche
I did that. I actually did it and I created it and then I sent it. People who were like, what do you mean? I was like, I did a whole 22 minute segment with an avatar in a podcast studio.
- [17:52]
Michael Stelzner
So does this mean that you can, for example, let's say I've got, before I started doing video podcasts, I've got a popular podcast episode I did years ago. Does that mean I could load that audio up right now? And the technology is smart enough as long as it's trained to use my actual vocals, but have the video model it or is it transcribing and redoing it? I'm just curious fundamentally how that works. Does that make sense?
- [18:18]
Aoife Roche
Yeah, so it would depend on how you'd want to do it. So the first thing that we would do is create the visual, right? So we would have the visual there. Let's say we took a visual of you right now with the headphones on and the gray T shirt headset. So we'd create the visual, then we would take that audio clip and we would either you could upload it directly onto the platform and it would create the entire episode if you took the full script, or you could take that full script and then put that into an A GPT assistant that was to look for viral content, find a couple, let's say any under the 90 second marks of like these bite sized pieces, pieces of content and really release those as footage. Or even better still, you could do both. And then you can do the long form content, publish that on YouTube, then do the short form content, load up the videos and take the short form content and redirect them over to your YouTube and you've still never recorded the video.
- [19:10]
Michael Stelzner
So just to be crystal clear, what we're really doing here though is we're Taking a transcript of my audio and then re rendering it in the new AI audio. Is that correct? So it won't sound exactly the same.
- [19:22]
Aoife Roche
Correct.
- [19:22]
Michael Stelzner
Okay, cool.
- [19:23]
Aoife Roche
Well, just on that, you can upload the actual audio too. You don't have to. Yeah, no, we have. We support the uploaded the audio, so.
- [19:31]
Michael Stelzner
So the actual audio and then the video will track to my audio and my breaks and all that stuff and it'll sync.
- [19:37]
Aoife Roche
So yeah, you can upload the audio file now. We do find it's not as perfect because I have a challenge with my accent. Because of the Irish accent. We don't have enough data because we're powered by 11 labs, which is the market leader right now in the voice cloning technology. But they just don't have enough of a data set on the Irish accent yet. So I sound a cross between British and Australian when I do it.
- [20:00]
Michael Stelzner
Got it.
- [20:00]
Aoife Roche
We used to have to do that. I had to load up my own audio file so that it would be able to get my accent.
- [20:05]
Michael Stelzner
Yeah. We're going to talk folks, about the tech that is powering this behind the scenes if somebody is brave enough to go ahead and try this on their own. Now, other things we talked about when we were prepping for this interview, outside of podcasts and reels and shorts and stuff, is also welcome videos. And then you also said something about interactive video. Is that what we were talking about earlier with the example of being able to interact with somebody kind of like a salesperson or something like that?
- [20:29]
Aoife Roche
Yeah. You can have interactive video elements on your website for when customers ask questions. You can also have your FAQs on your website. Could then, you know, you take the answer to the FAQ so that you create a video section on your website. We have one that someone's testing right now at the moment with MMS marketing. We're using the avatar videos to create videos and then we're sending those out as mass text message campaigns where people are getting a video now instead of an actual just a little short form text. You can use it for the welcome videos.
- [21:02]
Michael Stelzner
Webinars, you probably could use them for webinars, I would imagine maybe.
- [21:05]
Aoife Roche
Yeah, especially if you've used the interactive. Right. You had a moderated zoom there and someone was typing into the chat and the chat is connected into the interactive. And I think we will get to that point. And you know, they're trying to find how it's. Because the challenge that you have is the like I don't know if you saw there at the end of December, they came out with the legislation saying it's now illegal to buy followers. Like some views on social media. Right. You know, it's a violation of the guidelines to be able to do that from a business perspective. But I'm like, yes, that would have been useful five years ago.
- [21:37]
Michael Stelzner
Oh yeah, for sure.
- [21:38]
Aoife Roche
But now everybody, everybody's already done it. So you know, the horse is bolted now we're closing the stable door. But so I think that it's very much the same with AI. They're trying to control it, to figure out how to control it between. Because if we have a streaming avatar, right. Let's say I'm using a stream and it is there and people are working to try to make it better. Like you don't know if we're having this zoom. Are you talking to me? Are you talking to my avatar? Because if my avatar is pre trained on a data set in the background to, on my data set it could speak like me, talk like me, look like me, respond like me, but it may not be me.
- [22:15]
Michael Stelzner
That's where disclosure is probably important, right?
- [22:17]
Aoife Roche
Very. Yeah. I always try to, when I send out, I'm like, hi, I'm Avatar Bill or this is Avatar EFA or. But the reality is we do have several profiles. I won't name the platforms. I don't get people in trouble that are 100% run by Avatar versions of my clients.
- [22:37]
Michael Stelzner
Well, and frankly, just as of a few days ago, Facebook recently rolled out some new AI stuff that they're doing. They're creating these AI characters, but it's pretty obvious they're AI. But it's fascinating, fascinating. You probably already know this, but marketing is changing faster than ever and AI, it's leading the charge. At Social Media Marketing World 2025, we're bringing together the brightest minds in AI marketing to help you stay ahead. We're talking real strategies, proven workflows and hands on training you won't find anywhere else. Our AI track is designed specifically for marketers just like you. Ready to transform your marketing. Join us in San Diego this March. Head to social mediamarketingworld.info and secure your spot today. What do we need to be thinking about before we. Let's say. All right, ifa, you've kind of sold me a little bit on this, but there's some things maybe we need to process in our mind before we actually begin with something like this.
- [23:37]
Aoife Roche
Well, definitely. And I've had it more than once when people calling me in an absolute state of panic because you have third party vendors and suppliers or whatever. I'm not the only person who has access to this technology. We're just powered by a technology. And they're like, I signed up, you know, something has gone wrong. Now this person has my visual, now they have my voice. I don't know how to get it back, I don't know how to get it off. So I need people to really understand. You have to be very mindful of the ethics of the people that you're dealing with and understand what it determines and conditions that actually are in place for you. Because when you work with me, or it could be any of the other tech partners who are leveraging it, I literally have the power to create a video of you saying absolutely anything. Right? So you've got to be careful with that about now. I'm ethical. You know, we have very strict guidelines around what we use and what we don't use because we do a managed service. But I could literally do a video and say, hey, I think we should give Aoife all of our money. Let's pay her millions of dollars for these videos. And it would look like you, it would sound like you, and to the naked eye, people would think that it's you. So I want people make sure that you are reading the terms and conditions before you sign up. What is the opt out policy? What actually happens if for us, our avatars expire and you cannot re enable them unless the active subscription that was originally that created the thing is reactivated. So once it expires, it's toast on our side. Right. So we can't ever reuse the avatar. So I think people need to be very mindful of the platforms that they're using and who they're dealing with. And then the other thing that I see, I wanted to do this in the beginning, but I didn't do it for this reason. And now I see a bunch of other companies doing it, but I'm still not the ethical side of it. I'm a little bit on the fence about right now. I'm seeing a lot of people sign away their visuals, you know, because, you know, like actors and people who are like, maybe not trying to get ahead and it's just not working. So then they get the opportunities and they go into these big. Because they do run big events for recruitment events. And you go in and you just film your two minute video and it seems like nothing, and it sounds like nothing, and then you sign your little waiver and you walk out the door, but you're essentially signing your visual away. And that visual is Going to be sold to somebody else. So if you're someone who's like a devout vegan, you can't be upset if you end up on a McDonald's because you signed away your visual. Right. So people just got to be careful to understand. Even though the technology sounds as simple in theory, to create it, you've got to be aware what would happen if that fell into the wrong hands.
- [26:20]
Michael Stelzner
Awesome. Okay, well, let's say we want to get started. You mentioned multiple times that we need a two minute video. Talk to me a little bit about. I've seen some bad iterations of people messing around with hey, Jen. Where the video just looks a little wonky. Something's wrong, wrong. Right. So how do we create a video in such a way that is going to increase the likelihood that it represents me correctly on the first go round?
- [26:45]
Aoife Roche
Yeah. And that is the essential part. Right. Because the output is only as ever, good as the input. Right. And it surprises me how much people want to shortcut this process. And then they get upset when they have an avatar and it comes out and it's like you've given me a foggy lens or you're zoomed into the camera or, you know, I'm like, like, we all have 4K cameras. Nearly everyone has a 4K camera. Just you have to have good lighting, number one. Right. Sounds basic. But you would be surprised the movements. Do not try to script something and read it. Okay. Because then it's like a focused effort where I can see someone's really concentrating on what they're trying to do. What I talk to people when we record the video. I'm like, tell me about your business. Why do you do what you do? Very simple, should be very conversational. No big hand gestures. So if I did this in the video because I told you the way the technology works, it rolls back and it rolls forward. So if I do this at 30 seconds in, you can be pretty guaranteed. I might be doing a video talking about a really serious topic, and all of a sudden I'm going to go like this. And it doesn't fit into the video. So you got to be very careful with your hand movement, eye contact. I don't think people often realize how much their eyes flicker, you know, so I'm like, no, we do have technology now that fixes the eyes, but I think it can make people look a little bit odd. So I prefer not to do that.
- [28:07]
Michael Stelzner
So look directly into the camera when you're talking, obviously.
- [28:10]
Aoife Roche
Yeah. Don't look down, don't look up. Laughing, that's another one. People tend to get nervous and they'll laugh, you know, when they're doing it. And I'm like, yeah, but that's okay. But that laugh is going to come into a video and it may come in at a very appropriate time. We try when we're onboarding people, we have calls with them and, you know, we give them and the instructions do that. We very rarely have someone get their avatar in the first attempt.
- [28:35]
Michael Stelzner
There's a lot of things like I'm watching you as you speak, and not everybody's watching the video version of this, but there's things that you naturally do. Like your eyes get wide sometimes when you're excited. You look up to the left or the right, depending on when you're thinking and you nod your head from side to side and you do all these different things. Right. When you're listening to me, I would imagine it's kind of important you not be super stiff. Stiff, right.
- [28:55]
Aoife Roche
For sure.
- [28:56]
Michael Stelzner
You want to actually be as natural and comfortable as possible because it's gonna hopefully pick up your natural nuances. Like you said you weren't aware of yours until after you really trained the model. And you're like, oh, when I blink, I blink a little longer than your average human, which I never would have picked up on. You know what I mean? And smiling, right? Like, if you want to smile, I would imagine you're gonna have to smile if you want the avatar to smile. Right?
- [29:18]
Aoife Roche
For sure. And people take it very seriously, you know, And I'm like, it's almost. The more relaxed you are, the better the avatar. Because as long as there's not big, bold movements that's going to throw off the avatar. Like, I'm not the biggest fan unless it's a walking avatar. I'm not the biggest fan of a.
- [29:39]
Michael Stelzner
Standing avatar because your shoulders move around too much or something.
- [29:42]
Aoife Roche
I think people just kind of. When you put a camera and you get someone to stand in a spot and then you put it in their face, I think they just inherently get nervous. You know that where. When they're seated or. Because you can do walking avatars now as well, right? So the avatar is moving.
- [29:56]
Michael Stelzner
Okay.
- [29:57]
Aoife Roche
You can actually film your two minute video of you actually moving and it'll. It'll track and trace that movement, which is pretty cool. But I like to have people sitting down either in a chair or at their desk or something kind of conversational, and I think that creates a better avatar. I've a couple that I did in front of green Screens. We spent all this money getting them done when we were just learning. And I'm like, that's not how I move or how I talk or. So I think that, you know, the simpler you can make it, the better.
- [30:26]
Michael Stelzner
Does it have to be two minutes? Can it be longer than two minutes?
- [30:29]
Aoife Roche
We do. Longer is better in many cases because you get more kind of movements and gestures. But yeah, minimum is two minutes. But it surprises me how people, when it comes to doing a two minute video, talking about themselves, they're like, you know, it's a long time. It can come across as a long time, but it's really not.
- [30:46]
Michael Stelzner
Okay. So for people like you who have longer hair, talk to me about how it handles the hair movement and all that kind of stuff. Right. Because that's got to be an issue sometimes, right?
- [30:55]
Aoife Roche
I've had the issue with the beard.
- [30:57]
Michael Stelzner
Oh, really?
- [30:58]
Aoife Roche
Yeah, yeah, yeah. The technology. And it's gotten better with the last couple of iterations. But the beard moved okay.
- [31:06]
Michael Stelzner
Oh, but, but the mouth didn't move and the beard did move okay, what about like, does it get. You have long hair and when you nod your hair moves a little bit. Does it capture all that stuff?
- [31:15]
Aoife Roche
Yeah, it'll move. We have one which is someone had his hand on his lap and he was tapping his fingers like this. And the tapping of the fingers is in. So which actually makes it look more realistic because you're like, okay, you have this hand movement here, which is just seems very natural. You couldn't do that with AI, but it is, yeah. So gentle, natural kind of things are good. But there are some nuances. Like the beard one was a problem for us, but we haven't solved that yet.
- [31:44]
Michael Stelzner
How much of you is in the frame typically when you're recording? Are we talking like chest up or what part are we talking?
- [31:49]
Aoife Roche
Yeah, so I always say, I always record horizontal. Okay. Because you can always cut into vertical, but you cannot take vertical into horizontal. So we always get people to. And then our software, the way that I'll do it, it'll just take it and it'll cut into the vertical video. But don't be on top of the camera either. Right. They send me these videos and then we cut it into vertical and it's like there's like massive of zoom in. And I'm like, that's not helping anyone. You know, let's say. So you want to be like, maybe, you know, if I was to do one maybe like two feet back kind of camera so that you have a better range to play with because knowing that you're going to probably cut the video.
- [32:28]
Michael Stelzner
What about the backdrops?
- [32:29]
Aoife Roche
It depends on how you want it. Like, you can put your green screens because we do a managed service. We strip out the backgrounds and put in because we do fully edited videos. But if you were like, let's say, for example, if you do your content on a regular basis in like, let's say what you do and in the podcast studio, then that's your natural environment. I'm going to put in your natural environment because then it's going to look more natural. But you can always put it in front of a green screen. There is functionality now to remove the backgrounds. There's also functionality to change the clothes. But I'm not sure if we're really at the remove background because, you know, sometimes there can be something. If someone have something green, even on green screens. Right. With the best edits, you can still have things blurring in that shouldn't. So I think we'll get there eventually. But I would just say do it in a natural environment.
- [33:21]
Michael Stelzner
Okay, so let's talk tech for a second here. Once we've recorded a really great video and we've listened to all the tips you're talking about, let's talk about Heygen and 11 labs a little bit. Like, for those that want to try to do this on their own, tell us what we need to know about each of these tools.
- [33:37]
Aoife Roche
Yeah, so heygen is our tech partner for the visual. Right. And they are, I would say, the best, you know, people to have other, I think there's other competing products, but they are the best visual technology for sure. With them you can just sign up. It's very, very straightforward to go in and create your avatar. They give you the brief instructions and you do that. Now the only challenge that I would say on that, like it's the. The platform can be a little bit complicated to learn and to use. There is a little bit of time that you need to put in there to actually learn the platform. And especially if you have scenes or scripts that run over multiple variations and things like that, it can be a little bit of a challenge. So there is a little bit of work there, but they do have their help section in their videos, but they don't offer anything that's a managed service. You're on your own. It's a completely do it yourself real quick on.
- [34:28]
Michael Stelzner
Hey, Jen, does it only do video and not audio?
- [34:31]
Aoife Roche
You can't extract the audio independently of the video.
- [34:34]
Michael Stelzner
You can or you cannot.
- [34:35]
Aoife Roche
You cannot.
- [34:36]
Michael Stelzner
Okay, so is that where 11 labs.
- [34:38]
Aoife Roche
Comes in, you can upload audio to heygin. So let's say, for example, for anyone who's not familiar with ElevenLabs, Eleven Labs is simply a voice cloning technology. Right. You can upload two minutes of your voice up into 11 labs. It'll take it up, store it up there. And then you can synchronize your elevenlabs using an API into heygen.
- [34:57]
Michael Stelzner
Okay.
- [34:57]
Aoife Roche
You can have multiple different. So the way it clones your voice on the agent side, similar to what we do, you either extract the audio from the video, that video that came in, or you either upload your own audio file or you have your elevenlabs account where you've created multiple different audio files. So you've kind of three ways to get the voice into the platform, but using the API in ElevenLabs. And it also gives you more control in ElevenLabs, which just tonation and cadence and accuracy. But you might have a very excited voice versus you might have one that's a little bit more vanilla. If you were doing your course and we put that into super dupe where you could adjust the tonation because we realized it's not the same for like how I do a TikTok is not the same how as I record a course. Yeah. You can either try to do that in 11 labs and import it that way or. But hey Gen in general will just extract track the audio from the video.
- [35:54]
Michael Stelzner
Okay. So I think what I hear you say is that 11 Labs is really the superior voice cloning technology. And if you want to have the most amount of flexibility. It sounds like you're going into 11 labs and you're training 11 labs just on your audio of your voice in the same way you train hey Gen on your audio and your video together. Right. And 11 labs is hey gen even powered by 11 labs. Or do you think that's okay? So hey Gen is actually utilizing 11 labs, but they don't have all the features and functionalities. So you can go directly to 11 labs is what I'm hearing you say and really tweak your audio so it sounds exactly like you. Does it even get things like breath and just the little isms that we have in our voices, you know how we have like different things that we do in our voice.
- [36:40]
Aoife Roche
Yeah. Still not perfect. Right? They still work on 11 labs. Like we. We record audiobooks in 11 labs right now. And like sometimes it can go up and down or they can be, you know, something will change. Or one paragraph was on perfect and another paragraph doesn't sound perfect, but that's the evolution of AI, right? And you have to say today is the worst it's ever going to be.
- [37:02]
Michael Stelzner
Got it.
- [37:02]
Aoife Roche
But yeah. So then you would connect your, you could API your 11 labs account into hey Gen and then you get full access to the suite of voices that you've been able to adjust and change in 11 labs, where in Heygen you'll just actually get the voice. Like it has one voice that it took from the video, unless you upload a different voice.
- [37:23]
Michael Stelzner
Got it. So if you really care about your voice, and you might if you're a podcaster, because maybe you have a big library of incredibly high quality audio that you've created, you could train up a really good version of your audio inside of ElevenLabs, connect on the backside via the API to that, and then heygen. Is heygen smart enough to know when the tonation changes or the excitement level changes to change the person's. You know how like when I get excited, like my eyebrows go up and my facial characteristics change? Is it smart enough to know to do that based on.
- [37:59]
Aoife Roche
It's getting there. Okay, yeah, it is getting there. Where I'm seeing the, the gestures and the movements are much more succinct with the actual words being said. And I know that there was on their roadmap the idea that you would be able to manipulate the gestures so that you could actually make the. Now I don't think that's been released, but I know that they had that on a roadmap. So like for example, if I recorded a video and had a very excited movement, the idea that I would be able to screen grab that movement and be able to put it in and embed it into a video where I thought it was most appropriate. But we're not there yet. But you know, we'll get there.
- [38:37]
Michael Stelzner
What about editing? Is it to the point where you can. I get the sense it's kind of like you. You render it. Maybe you render a couple iterations, you get to pick your favorite. But is there a lot of editing capabilities in hey Gen where maybe you did something you don't like and you don't want that. You want that one little thing tweaked.
- [38:54]
Aoife Roche
Or is that you gotta re render again? Yeah, but they did move to putting it onto an unlimited package and I think that was to pretty much try to solve that problem. Right. Because you know, people go in and they try it and then especially if you've spent a lot of time rendering a video, right. And you've got it ready and then you render it and then you realize that there's a mispronunciation of a word or something. Like before, if that was like let's say a five minute video, you were charged five credits and then you found out there was a problem after the rendering and then you had to go back and get charged another five credits. And I think that there was probably some levels of frustration from a user perspective. So now they've made it unlimited. So you get unlimited attempts to try to correct it. But where the, the problem where most people will go wrong is the input video hasn't been. There's eye flickering or big hand gestures or something. That's one challenge. Or it will be that they didn't actually take the time to really read. Because people, it's like, this happened to me yesterday, someone sent me a script and they're like, oh, it's an amazing script. And I'm not taking away from the fact that the script was a good script, but it didn't read well.
- [40:03]
Michael Stelzner
That's what I want to talk to you about next. How do we actually script it in a way that sounds like it's speaking? Do you have any tips on that?
- [40:10]
Aoife Roche
Yeah, you have to listen. They ask me the exact same question. They're like, but I just want to send you the script. So you know, how do I make the script good? Because you can adjust and put tonation and punctuation. You can do that. And we put that in the super dupe platform where we have an entire window where you listen to it paragraph by paragraph before we ever allow you to go on to append it onto the visual. Because. And I sit there and I just listen. I almost close my eyes and I'm like, okay, if this was a conversation is this. And if it's not, then I'll go back and I might add a double exclamation mark or a space or I might separate the paragraph or I might have to phonetically have a word adjusted because like, for example, I had one. Their surname was F O R I E D and it's pronounced freedom, but it was calling a fried. And it's not wrong, but it's not correct either.
- [41:03]
Michael Stelzner
You have to spell it F R E E D or something like that.
- [41:06]
Aoife Roche
Yeah, exactly. So sometimes you have to have a play around with that. But people can just get a little bit excited and go, I'm just going to render the video and it's going to be perfect. No, you need that auditory aspect of it is the definition Between a really good video and just an average video. It's that cadence, it's the tonation, it's the pronunciation, and just knowing, okay, that word sounded good in a script, but it doesn't fit into this conversation.
- [41:36]
Michael Stelzner
Is there any, like, tools like ChatGPT or Claude that you can feed the scripted to and say, recraft the script in a speaking style? I mean, are they there yet? Do you understand what I'm saying?
- [41:47]
Aoife Roche
So, yeah, I try to do that with it and with. Especially with ChatGPT, but it's. It often will give you back the whole scenes and everything else. And then. So then I try to do it with only the narration. But you would have to do only narration with a trained GPT assistant that would be trained on your brand voice, right? Yeah, yeah, that would really be the way that you want to do it. But we're still the two things that I have right now there, the viral clips, being able to identify the viral. I think we talked about that. The viral pieces of content and the ability to read an audio in a way that it doesn't sound like it's AI generated. They're the two kind of things that are on my plate at the moment to try to figure out.
- [42:28]
Michael Stelzner
So a lot of people that are listening right now are like, man, that's a lot of work. I have to create a really perfect video, right? I have to train up, ideally, some sort of a tool like 11 labs on getting my vocal cadence style isms all the things down. Then I've got to make sure that I script it in such a way that it sounds like me. For some people, they're like, I may as well just do it. But for all the reasons you said earlier, people don't do this. The good news is, is I would imagine is if they were to work with an agency like you and what you're doing at Super Dupe, you're going to get them at least out of the gate with really good stuff. And then if they wanted to, eventually, they could probably go to 11 labs and train up their own model and go to hey Gen if they really wanted to. But what I love about what you're doing is you're bringing the video side of it, the audio side of it, and the scripting side of it all together in a way that's kind of challenging. And I'm guessing that once you get this figured out, then you learn to write in a way that you speak and then it just starts rolling. Is that right?
- [43:37]
Aoife Roche
Yeah. So, like, for example, with my Clients. And a lot of my clients are like, you know, CEO, CEOs. And you know, the last thing that they want to be doing is sitting down doing videos. Right? They just don't want to do that. Right. So, but what we do is we go, we'll take them in, we'll set them up, we work with them to create the visual and the video. That can be multiple iterations. We often take their voice because their audio, their mics. We could put it into Adobe Podcast, clean up the audio, do all of these things so that before we'll ever load it onto the platform, then we'll go in the background, we'll create a GPT assistant for them, we'll train it on their voice, their model, their brand, brand style. And then that GPT assistant is actually what's writing the script. So there's a lot of pre work done before we've ever even rendered a video. Then when we render the video, we take the videos then. And we've already taken. We've got like five of their top creators. We've got their editing style, their B roll style, their captions. So my clients end up at the beginning or end of every month either. On the very end of the Beginning, they get 15 or 20 videos that are just dropped into their Google Drive if we're not the one scheduling, scheduling the content. So then they're like, okay, this is amazing because now we tick the box of be present on social media in January, you know, and it just gets shopped into their inbox. So people will pay for that ease of use. And then where we schedule them for them, then they just, you know, I've had multiple people come back and say to me, I had to got so many comments about that video, I had to go back and see what it said.
- [45:06]
Michael Stelzner
You definitely have gotten me to the point where I'm like, hmm, maybe I'll consider something like this. Ifa, if people want to connect with you on the socials, what's your preferred platform? And be sure to spell your name out. And then people want to work with you, where do you want to send them?
- [45:20]
Aoife Roche
LinkedIn is one of my primary platforms. So that's IFA, that's a O I F E and it's roach or O C H e. For my LinkedIn platform, you can find me on Instagram as Well, which is iFads. A, Y, F E, J R, O, C H e. But yeah, LinkedIn is definitely the easier one. And then for it to actually work, you can actually go on to super dupe AI or just search super dupe platform on Instagram if you're curious to be like, actually I want to see what these avatars look like. Every video on that super dupe platform Instagram is an avatar so you can go and you can see all of the different ways and the styles so that people can actually because sometimes seeing is blue.
- [46:00]
Michael Stelzner
IFA Roche thank you so much for sharing all your insights with us today.
- [46:03]
Aoife Roche
I appreciate you and thank you so much for having me and I look forward to seeing your avatar.
- [46:07]
Michael Stelzner
Hey, if you missed anything, we took all the notes for you over@socialmediaexaminer.com a38 and be sure to follow the show on whatever podcast app you're listening to so you don't miss any of our future content. If you've been a longtime listener, would you give us a review on whatever platform you're listening on? And do check out our other shows, the Social Media Marketing Podcast and the Social Media Marketing Talk Show. This brings us to the end of the AI Explored podcast. I'm your host, Michael Stelzner. I'll be back with you next week. I hope you make the best out of your day and may AI help you become more successful. The AI Explored Podcast is a production of Social Media Examiner. If you're serious about learning more about AI and marketing, I'll see you at Social Media Marketing World 2025. Go to social mediamarketingworld.info and secure your spot today.