Transcription
Have you seen an AI avatar presented video like this one or this one? You can tell them a mile off, right?
If you've watched a recently made in-house training video, you could well have come across one, possibly without even knowing it. You might also have seen one on social media. You're also watching one right now. This video isn't really me, but an AI representation of me I generated through an application called Hey Gen. It's my digital twin if you like. Although this is really me. This isn't though. What's the difference? You can't really tell, right?
I previously produced a few different avatar versions of myself. Initially, I was sharing these on social media to gauge the reaction. This is my latest AI avatar. All things considered, it probably cost me about £1,000 to make. I very nearly set up a business with the intention of helping people create their own avatars. I thought they'd be huge and I still think that when they're used well, they can be an amazing tool, but people haven't quite gotten over their fear of being impersonated, and they are a bit fiddly to set up as well.
So, in this video, I'm going to show you how to create the perfect AI avatar. You'll have a digital twin that you can use to present, which can save you a ton of time if you want to make more video. There are loads of different avatar applications available. An article published in April by Justin Moore from Anderson Horovitz goes into some detail the current state of play for video avatars and provides an overview of the market. A link to the article is in the video description. Here are some of the different avatar tools Justin identified. So, there's quite a few, but for this video, I'm going to focus on Hey Gen. You can also clone your voice using 11 Labs. And then you can automate the production process through make.com using Google Docs for your script. I'll get into those details later in the video.
The main reason why avatars haven't gone big quite yet is because of what's known as the uncanny valley. Basically, uncanny means unsettling or mysterious, like this humanoid robot. It's meant to be a representation of a person, but it comes across just a bit weird. People think that with Avatar videos, too. Basically, if they're done badly, they can come across a bit robotic and weird. Now, a couple of years ago, the tech wasn't really there to create a super realistic avatar version of yourself. So, that's what was meant by the uncanny valley.
If you think of human interaction, so much is learned through someone's emotions just through watching their face. This is actually probably the number one thing we notice in human communication. Even before we speak, the face shows a lot of subtlety, and the best Hollywood actors are masters of this. It's a pretty hard thing for AI to get right, but I think we've pretty much gotten out of Uncanny Valley in the last year or so. The AI is good enough for realistic avatars now. Basically, I know this because I've tested it out a lot. I use avatars on my social media and some people have asked if it's really me or it's an avatar version. So, people can't always tell the difference with a good setup. But the longer your avatar video runs for, the more likely your audience is going to notice something uncanny. So I usually recommend using them for short videos or for business use cases where your audience is going to be small and you can let them know that your video is presented by your avatar. So I don't really recommend using an avatar for YouTube videos of say more than 5 minutes in length. YouTube exposes you to a potentially big audience and it's really a personal platform where you're trying to build up a close connection with that audience. you run the risk of being in the uncanny valley. And on YouTube with loads of other things your audience can watch, you just don't really want that.
But there are a lot of great use cases for an avatar, particularly in business. Let's say that you want to present training videos in your company over a couple of weeks. Using an avatar is a great method for doing that because setting up in an office multiple times can be tricky. You could also use the videos for sales material. Most sales messages are just text, but using an avatar enables fast video creation, which can really help you stand out. If you're a senior member of a team, like a director or a CEO, then you're more likely to give updates to your company, but you're going to be time poor. So, an avatar video can be used as a substitute for internal communication and business updates. Lastly, in my experience of making videos, something can go wrong during the shoot. You might have a clip where you got distracted or just messed up, but didn't rectify it when filming. This is where an avatar can help. So long as your video is shot in the same setup and clothing as your avatar, then you could clip in an avatar for a small part, then revert to the main video once you've made the fix. No re-shoot necessary. The main thing you're trying to save with an avatar is time. The real benefit is that once you've got a good setup and an avatar, you won't ever have to set up again if you're comfortable with your look. So using an avatar is just a great way to save time in all these use cases.
Before we get into the details of the avatar generating process, we need to think about setup. I've been through a few versions of creating an avatar, and the quality of what you get out of a platform really depends on your setup. So, let's start here. I did an online course called On-screen Authority to learn about the best setup for creating YouTube videos. And if you're a serious creator in need of some pointers for your setup, then I really recommend it. That's linked in the description, too. It's technically very easy to create an avatar version of yourself. And as a tester, you could just do it on your webcam or mobile, but I want to show you how I got from something like this, which is just on mobile without any lighting, to this, which is just a much better avatar and way more convincing. The main thing I got from the course is composition and lighting setup. These are hands down the most important things for creating a quality avatar.
So, first, let's think about composition. You need a reasonably big room with a nice background. I do most of my videos for my study now as I like the background and I can create a good composition. I'm sat in the middle of the frame, but I also line up to this white door in the background. I don't recommend using ring lights for your setup. They direct light onto your face and they'll also reflect rings into your eyes. You don't want that. If you just want to get started, sure, use a ring light, but there are much better options. By far the best investment is a round newer LED front light. These come at about $200 to $250. It's really good investment. The link is in the description. With this, you want the light to be in front and to the side. Turn up the brightness of the lamp until you have light on one side of your face and some shadows on the other side. Having a bit of light on that cheek is where you want to be. The next thing to do is get backlight so you create an edge of the opposite side of the face to where the back light shines like this. This really makes you stand out from the background. You can get a fancy light for this, but I don't think you really need to. You could even use a bedroom lamp so long as you create this edge. Then you're fine. So that lighting setup and composition is basically all you need to have in a pretty good looking avatar setup.
Now, not everyone has the luxury of having the space in their house or office to put up a good lighting setup. You might just have a small space that doesn't exactly lend itself to having a nice composition. Well, there's a workound. I recommend getting a small green screen to attach to the back of your chair. It's not absolutely necessary, but it can help. You still want to use some lighting, but you might not have the space to use a newer circle lamp. Just try using whatever you can to get a similar effect. With this smaller setup, you can do your own filming and later you can remove the background in runway which is another AI tool. It costs about $15 to get access to this feature. If you know how to use keying or rotoscope on After Effects, you can opt for that route. You can then stick in any background you like. Give it a bit of blur and now you're in another setting. You can get a good setup wherever you are.
Now we can move on to the recording and creating the avatar. One cool thing about Hen is you can create your own avatar for free. Just when you export from the free version, you'll have a hen watermark and the maximum export size is 720p. This is all good if you want to just test out the process without paying, but not really good enough if you want to use an avatar for a couple of videos over 3 minutes in length. So, I'll demo the creator plan. Hey Jen comes in at $29 a month on the creator plan, but the main thing you're trying to save by using an avatar is time. And your time has a value. So with just a couple of videos, this subscription will pay for itself. It's worth noting though that if you're looking to make videos of over 5 minutes, you'll need to go on the team subscription, which is only $10 a month more.
Once you've logged in, you just want to go to create your own avatar. Click through here, and you'll go through the steps. You'll need to create a verification code. Then you need to provide 2 minutes worth of material. It's not a lot really, so it's a very easy process, but there are definitely some things to be aware of when you're recording your material. Firstly, and while this might sound kind of obvious, you definitely need a fixed camera. So, at the very least, you need some kind of tripod with a phone attachment to make your recording. Links for all the recommended products are in the description. To make sure you're centrally in shot and it's working in the composition, put your phone into the folder and use the front camera to visualize where you're positioned through looking at the screen, which should be facing you. Once you're comfortable with the positioning, I recommend turning the phone around and using the back camera. It has a better lens, but you'll still be able to shoot 4K footage on the front-facing camera if you have an iPhone 11 Plus or any other phone after this generation. It's also worth getting a vertical recording so you can use this for social media later. Pagen allows you to upload three different avatars on the creator plan.
For the recording, all you really need to do is speak to the camera for over 2 minutes. You don't need to read from a script or use an auto quue. You can literally talk to the camera in this ad lib fashion. Hello, my name is James and I live in England. I have a cat named Thumper and she's a very nice cat, etc. When recording, you want one continuous take for at least two minutes in the presenting style you want your avatar to have. If you move about, roll your eyes, or visibly acknowledge you've fluffed up, then these will be recorded. And if these are uploaded to Hey Genen, it will mimic such movements, leading to potentially weird outcomes. So, you just need to talk naturally and directly to the camera for just over 2 minutes. But, you don't need to be overly concerned about exactly what you say.
If you've never presented to camera before, there are definitely some things to look out for. So, here's some quick tips. Sit forward with your shoulders back. Avoid slouching or leaning backwards because you'll end up looking noticeably pear-shaped. Also, sit relatively still. There may be a temptation to sway because presenting feels unnatural. If you're wearing a dark color, ensure it's free from dust. This is particularly important for men like me because I've got a beard and you can get beard dandruff on your clothes. Yeah, it sucks. Wrap some tape around your hand and take any dust off your clothes. Make sure your clothes are ironed and make sure your collar particularly is in good order. You'll notice these creases a lot otherwise. Consider your hands. If you're waving them around a lot, then the avatar will replicate it. There's actually no need to wear a microphone when doing an avatar video recording. Although I suppose it could make it more realistic.
A 2minute 4K recording is likely to be a couple of gigabytes. This will take some time to upload onto a system like iCloud. On an iPhone, the quickest way to resolve this is to go to your photo media library, click the share button, and save to files. You'll then go into iCloud, save files quickly. You can then navigate to iCloud on your browser and download the video to desktop. All being well, this is the file you'll use to upload into Hen. You can now go to Hen and upload your file. It's important you do this yourself because you need to provide personal consent through providing a recording. Once uploaded, the avatar will take a couple of minutes to process. It's surprisingly quick.
When it's ready, you can go into the video creator and write in your script. Hen's voice capture isn't the best, and you've only recorded 2 minutes of audio for the AI to train on, so it won't sound exactly like you. This is where you can go back into the uncanny valley. So, there's a couple of ways around this. The first is to just record your voice separately and then upload an audio file. I recommend getting a good microphone for this like a road NT USB which will cost about £100 or dollars. It is possible to use a lower spec microphone such as phone headphones with an attached mic but it's possible you'll get background noise which you'll need to clean up. It's just not the most optimal route when we're trying to save time. I like recording my voice for Avatar videos because it makes my voice sound the most natural, which means I'm less likely to be in the uncanny valley.
But you can't fully automate the process this way. So, I need to show you how to use 11 Labs. The first thing you need before creating a voice clone is quite a lot of audio pre-recorded so the AI can train on it. 11 Labs needs at least 30 minutes, but 2 to 3 hours worth is recommended. If you don't have any clear and consistent recordings of your voice, then this is quite a lot of material and a big time investment. So, you might want to skip this step. But if you want to go through with it, then you could just do something like read a novel out loud into a microphone. But you need a good microphone like the road I mentioned, and you'll need to edit the audio file to remove any fluffs in the recording. As with the video avatar, what you put in is what you got out. So, if you leave in lots of repetition, ums, and Rs, then the AI is going to have more difficulty training on it. Worse, it might actually think this is a natural way to speak and insert weird pronunciations into the voice clone. You don't want that. And if you've already spent a couple of hours making a recording, I'll assume you're serious enough to clean it up. Audacity is a free tool you can use for the editing step. Or if you'd rather just not do this, then find someone on Upwork to do it for you. It's not a very expensive process, and you should be able to get a clean version for about $30. You'll need to be on at least the creator plan to get a realistic voice clone, which is $22. Once you're in, it's pretty straightforward. Just follow the steps to upload your voice file, and the AI will do the rest. In a couple of hours, you'll have a voice clone. It's never going to be 100% perfect, but it can get pretty close. Your audience on these videos is largely going to be people who you don't know closely, so I don't think you need to worry too much. Then you can go to your text to speech editor and select your professionally cloned voice. type in a few words and play back. My AI voice is now speaking to me. Uncanny, yeah, a bit, but I'm fine with that.
Hen also integrates with the 11 Labs API. So, if we switch back to Hen, you can select your cloned 11 Labs voice in the app. The next part is to automate this process. The good news is you don't really need a complex automation to do this. I built a pretty straightforward two-step one in make. It just saves some time. This automation basically just takes your script, pushes that into 11 Labs and then into Hen. The second part of the automation saves the video file into Google Drive. In this first step, you just want to watch files in a folder. So, set up a folder in Google Drive called avatar videos where you can put your script. The Google Docs part is where you write your script. For that, you just need to map the file ID from what's been recently created in Google Drive. The automation only pushes to the next step when an approval code is written into your script document. In this case, it's approved 1 2 3 4 5. After that, 11 Labs didn't have an integration when I made the automation, but you can hook it up via that route. I just used a HTTP request to post to the API. Then it's pushed to HGN to make the avatar video. In the second step, I used a web hook to say when a hen video is completed, it's uploaded in a folder of your choice in Google Drive. It's not a very complex automation, so if you know your way around make, you'll make short work of this. I've also linked the JSON blueprints in the description below.
So, that's everything. This is my complete AI avatar, fully voice cloned version of me, telling you that this is the end of the video. If you follow this far, I expect to see your avatar hitting social media soon. If you like this and want to get more out of AI, then like the video, leave us a friendly comment, and make sure you subscribe to the channel. Also, send us a message with what you've created or if you need any other tips on setting this up. See you next time. It will be absolutely agentic.