WARNING: YouTube is Quietly Removing These Videos

So I just wanted to warn you to stop posting these types of videos to YouTube if you actually want to build a real big YouTube business right now. A lot of things have been changing on YouTube recently, and AI's been getting so good that it can produce a lot of types of videos incredibly easily. And I personally run two different AI companies. But what if I told you that a certain type of AI-generated content, as well as a certain type of content that got me millions of subscribers on YouTube, is content that I would not actually post? And I would highly recommend stopping posting these types of videos to YouTube and posting a different type of content instead. And I'll show you the better different type of content in just a second. But first of all, I made a video recently talking about the news that YouTube CEO said basically in more or less words that he's going all in on YouTube. He sent out a memo privately to the employees at YouTube. They've also recently integrated Google Veo DeepMind 3 directly into YouTube Shorts, and they're probably going to do the same for long form soon as well. And this presents a huge risk for us faceless creators on YouTube. And that risk is commoditization, which means when it's so easy to generate videos, then everybody can do it, and there is infinite competition. And when there's infinite competition, it makes it very hard to grow or to distinguish yourself. And AI is just going to make it easier and easier to generate these entertainment type of YouTube channels where it's just funny AI-generated visuals, or even channels like Hominid History Hub, which I recently went over here on this channel, and I showed that they got 90 million views within literally just a couple months doing these AI-generated videos, but when it's so easy for anybody to generate these types of videos, it'll create too much competition. So what's the solution? The solution, in my opinion, is building brand IP. Intellectual property, or IP, is the image or brand that you own that exists in other people's minds. And this is something that cannot as easily be commoditized by YouTube and by thousands of creators generating the same exact videos. And then even if they do, even if they copy you, it builds your brand and it builds your image. And no, I am not saying to start a personal channel. That's certainly an option. And I have a personal channel that you're watching right now, but most of my channels are completely faceless, yet they're still benefiting from building IP. So you might be like, Matt, how do you build IP with a faceless channel? Well, the secret is running channels like these two channels right here, which both get hundreds of thousands and the other channel gets over 2 million views per month. With faceless content that still has a host. So how is this possible? It's by using AI avatars. So Zinni Studio right here does a 3D AI avatar, and you could easily generate something similar using Seedream to generate the actual image and then using HeyGen to generate the lip syncing and the talking. That's one method, and it's the most expensive method. Another method is doing like the channel Julian Goldie SEO and making a hyper-realistic clone of yourself. And he gets 2 million views a month going over different AI tools and whatnot. Again, that's a very expensive method. A cheaper method that still gets millions of views per month in a completely different niche is the channel Yellow Dude. They do faceless tutorials for doing bodyweight exercises. They also get millions of views per month and they can easily generate characters like this, simple 2D animated characters doing different workouts using any AI image generator out there, whether it's Seedream through something like OpenArt, which is how I use it, or whether it's ChatGPT. You can generate these types of characters and then regenerate them doing different poses and different actions. And even if it's a super simple stickman animation channel like Warren Stick or like Money Stick, which are blowing up right now with simple stickman animation channels, this is what I would recommend. You're building real brand IP by doing this, and it's not as easily going to get commoditized. Now, what also do all of these channels that I just shared with you have in common? That gives them a huge advantage over all the other AI-generated content. Because you might be saying, Matt, you just showed me how to use AI in a different way. What also separates all of these channels right here? And why are these better? And why are these the best way of building a YouTube business right now? And the answer is that all of these channels teach something. Also avoid content that makes any huge claims. Do not post videos saying how to make $100 a day online or how to make a specific amount of money. I actually recently dealt on my channel due to making videos like that. I got 2 strikes on my channel, and if you get 3 strikes, your channel will actually be deleted. And so I've changed my entire content strategy of staying away from those types of videos and more so just doing complete guides and complete tutorials and not putting phrases like that in titles or in thumbnails, even if in the actual video, like I always make sure to do, you say, you know, obviously I can't guarantee you'll make any specific amount of money just by watching a video. Even if you say that, YouTube's automated systems might flag your video as being spam or breaking the YouTube guidelines, and you could lose your channel. So you need to be very careful about any claims that you make in your videos. Also, it helps putting a disclaimer in your descriptions as well if you give medical advice or money advice. But that's not enough. Make sure that your titles and your thumbnails don't promise anything specific, especially when it comes to making money. And channels like HowToAI were huge channels that got taken down completely due to this. So don't let that happen to you. And you might be saying, Okay, Matt, you told me what not to do. What should I actually do when it comes to generating videos the right way where my channel is going to be completely safe for the long term? Because there are channels that just do completely AI-generated visuals that get some views in the short term, but then their channel gets taken down as we've gone over. Now, I've been able to build about a dozen YouTube channels that I've been growing over the past 10 years, and I've added some of these channels along the way. And luckily, I haven't had a channel get taken down yet because I really try my best to follow the YouTube terms of service. But the The thing is, is that I am using AI on a lot of my channels. So you might be asking, how are you doing that? Well, the way that I'm doing it is that I'm not just using AI to make the same commoditized videos that everybody else will be doing. Now, in the video that I uploaded before this one, I showed how to use Claude Code to generate full, complete YouTube videos from scratch. And I saw a lot of comments on that video from people saying, Matt, this is awesome. However, isn't this the type of content that YouTube is going after, that YouTube doesn't want on their platform? And I've actually done some videos talking here on the channel about what YouTube is banning, what they're deleting, and so forth. And I filmed that last video largely to show what the technology is capable of, but I don't think I went as in-depth as I would have liked to have, showing how to customize it and how to actually do this correctly. So the first part when it comes to making incredibly good videos is your script. Your value is actually created within your script. And in the last video, I showed how to use Claude to just generate a complete script. Again, just showing proof of concept what it can do. It actually generated a full entire video. However, when it comes to your script, you definitely want to be making sure that it actually gives value and it follows all of the best practices for YouTube scripts. Now I have a software, it's called TubeMagic.com. It'll write you amazing scripts, but only if you use it correctly. You need to put in several context videos so it pulls the best parts of those videos out and then puts them into your script, and then TubeMagic will use all of the best templates for writing scripts and the best strategies for doing so in order to make an incredible video that is incredibly valuable. So now you got your script, you used AI a little bit to help you with that. Now, how do you actually generate the visuals and generate the full entire video using AI? Well, number one, don't just use the same AI voices that everybody else uses. If you're using something like ElevenLabs for AI voices, My recommendation would be either to clone your own voice or use a lesser used voice. Like, for example, my voice is on ElevenLabs if you'd like to use it. And so now you have your script, you have your voiceover. Now it's time to generate the visuals for your channel. I'm going to share with you a way of doing that that's a lot different than what most people teach. If you watch most YouTube videos about how to create a faceless channel, they're going to tell you to just stitch a bunch of videos together that you find on different sources on the web, whether it's from stock footage or an AI video generator. And then call it good. That is not what you want to do whatsoever. Yes, it is true. You can get clips from a lot of different places. For example, I use stock media assets. I also use the sponsor of this video, which is called Higgsfield AI, to generate clips for my different channels. A lot of my freelancers have access to my Higgsfield account. Again, big thanks to them for sponsoring this video. You can check out Higgsfield down at the link below, but that's still not enough because number one, you have to manually generate every single clip, which takes absolutely forever. You have to edit it all together and it still doesn't have that thing that separates you from all of the other channels out there. So what will separate you? Well, it's having a face of your channel. And that doesn't mean that you have to get on camera, although you certainly can. I've grown a big YouTube channel getting in front of a camera, which is the one that you're watching right now. However, I have a lot of channels where I never get in front of the camera, yet I still have a host who's the host of the channel. So how in the world am I able to do that? It's by using AI to generate a character and then using Claude code to generate the video and then implement the character into it. And I want to show you how to do that right now. And the way we're going to do that is that we're not just going to use Claude to generate a video because it can't generate a video on its own, but we're still going to use Claude, but we need to connect something to it. And what we're going to connect to it is the best AI platform for generating visuals right now, which is Higgsfield AI. And they actually created a custom MCP, which is basically just a way of connecting their tool with AI. So then you can automate this entire workflow. It'll generate the full video for you and you can have an avatar who's the face of your channel. And so right now I'm going to go over exactly how to do that. First things first, you need a face of the channel. And if we hop on over to my computer right here, in the last video that I did, I showed how I used Claude to generate a full, complete YouTube video. Right here. But the problem was that there was no face of the video. There's no character like I'm talking about. And the benefits of having a character is that it will not get commoditized because let's say someone steals your character, they're just building your brand. And if you have a name for the character, you have a personality for them and everything else, they're just building your brand by using it. And then you can use this character to sell products easier, have increased trust with your viewers. And there's so many other benefits as well. So in my opinion, it's a complete no-brainer. And so if you want to see exactly how to generate this full video that we did right here with Claude Code, then you can feel to watch the last video that I did. I show exactly how to set it up and everything. But just as a quick recap, if we go into Claude Code right here, which you can download from your computer just by going to Claude.ai and then going to download their desktop app, and then from there you're going to go to Claude Code right here, then you need to make sure to install the custom connectors. So if you go to connectors right here, you're going to see we have Higgsfield installed and we also have ElevenLabs. And I show in the past video exactly how to install them. But you can essentially just go to the Higgsfield MCP right here, copy this link right here, and then go ahead and paste that into Claude in the settings. And you can have these connectors. And in order to actually edit these videos together, then what you're going to want to do is have Claude do it for you by installing something called FFmpeg. Now I have a prompt sheet in the description of this video that includes the prompts for doing this, as well as that command for installing FFmpeg. Which will edit your videos together for you. And Claude will do all the editing for you in the backend. And if this sounds confusing, it really isn't. You only have to do this process once. Just follow step by step what I showed you in the past video and you'll have this set up in no time. And so as you can see, I ran that command right here. It installed FFmpeg and then I gave it my prompt. I gave it my voice ID as well for the ElevenLabs voice, and it's using my ElevenLabs account and my Higgsfield account in the background. To make this video, and it went ahead and made this entire video, which is awesome. However, like we said, there's no character. So here's what we're going to do. First things first, we're going to generate a character. In order to do that, I'm going to go to— go to Image, go to Generate Image, and then I'm going to use GPT Image 2 because in my opinion, it is the best out there right now. And we're going to do a slightly vertical shot for that and then just go 1K for the resolution and then go medium because it uses a bit less credits. And I think it's more than good enough. For this. Then since this video is about 5 things that hotels hide from people and is talking from the perspective of a hotel front desk agent, we're going to say to generate an animated picture of a hotel desk agent on a white background, likeable character. And we're going to actually change this to be a transparent background right there. And then we're going to do 4 options here. Just, it's nice to have a few and click generate. And now we've got a few options here. A lot of these look really good. This one looks really good right here. So we'll go ahead, right-click, copy that, and then drop it into Higgsfield's Background Remover. Because even though it shows it's transparent, you can still kind of see the background there. So then we'll click Generate for that, and this will actually remove the background fully. And so now we have that transparent image. So what we're going to do is go back on over to Claude. Now, if you wanted to have this character added from the complete beginning, if you're doing a complete new prompt from scratch, then again, you will find these Claude code prompts right here in the description of this video. And all you're going to do is then add to the bottom of this prompt what I'm about to show you right here. So we're going to first and foremost upload the picture of our guy on the transparent background. That's our character. And then we are going to give it some instructions. So since I've already generated this video in the past video that I recorded, and in this one I'm going to show you how to add it to an already finished video. But again, if you're doing this for a video from complete scratch, just use the prompt that I give you. And add the following section to the end of the prompt. And I'm going to give you that section in a second. But also, I just want to quickly say as well that you could generate a video, for example, using something like vid.ai, which is my AI generation software, and uses mostly stock footage and AI-generated clips to make full YouTube videos up to 20 minutes long. And then you could run that video through Claude Code in order to add a character to it. Now, it will do all of the animations and everything else completely for you. But if you wanted a character who's the face of the channel, which is going to help keep your channel safe so you can stop posting those videos that I recommended stopping earlier, this is the best way of doing that. So let's jump onto my computer right here and take a look. I'm going to add the following prompt: use the video that you outputted and add this character as the host of the channel. Make him talking and also have 'Occasional clips of him with lip-sync talking to the camera in a hotel.' And of course, you're going to make this custom to whatever your channel is about. So if it's a channel about swimming lessons, they might be in a pool. And for the rest of the video, just put him small in the bottom right corner of the video. While it's going. Same voiceover, same video, just adding these few tweaks to it. And then also even just to give a little bit more value, we're going to say also add some captions for the first 10 seconds of the video, since the beginning of your video is going to be the most important part of it. So adding things like captions are going to help keep people's attention. In the beginning of the video. Then we're going to again, make sure if we click this plus button and then go to connectors, make sure we have Higgsfield turned on and the ElevenLabs player. This will use our Higgsfield account that we connected as I showed you how to do in the past video. And it'll use ElevenLabs, which I also showed you how to connect in the last video in order to generate that right there. Then we'll go ahead and click enter. And now Claude Code is going to do all that work that we just asked it for us. Now, again, if you're doing this from the complete beginning, like starting up with this prompt, you're just going to add a variation of the prompt that we just did. So we're going to say, and use the attached character as the face of the channel, have them in the corner of the video as well as popping up occasionally throughout the video as well with lip-synced audio. So the character is actually talking and then Claude Code will do all of the work for us. So Claude's thinking right now in the background. This is going to take a couple of minutes. And just to walk through what Claude's doing right here, it's uploading the character to Higgsfield. Extracting 3 audio snippets from the voiceover, generating 3 different lip sync talking head clips using Juan 2.7. So it even found us the best lip sync model available in order to do this. And then it's using FFmpeg to compose this all together, plus putting the character P'Nchi as a pop-up in the corner for the rest of the time, withdrawing text captions for the first 10%. Literally what we asked it exactly to do, and it can do it all for us. Now, one thing to note while we're doing this, I didn't talk about this enough in the past video. Is that we do have to click allow for a lot of stuff. Like we have to allow that right there. And there's generally going to be several pop-ups that we have to allow. But once all of these permissions are allowed, we don't have to do it again. So for generating another video with Claude Code, we don't have to keep doing those same things again. And since I already gave it the permissions for this past video and having to click allow a bunch of times, I don't have to do it on this one. And now while that's generating, a lot of people were asking in the last video, how much did it cost to generate that full complete YouTube video using Claude? Now it was a couple of minutes long. We used some See Dance clips in it. But if we jump on over to my computer, as you can see in the past 7 days, this is counting some of the stuff that we're generating right now. But in total is probably around $6 to $8 for a 3-minute video. And as you can see, the majority of that is coming from Seedance 2.0, which is the most expensive model, even though Higgsfield is the cheapest place to use it. And then the rest is from VO3.1, which is slightly cheaper. And then GPT Image 2 right there. So the total cost including everything would probably be around $18 or $20. For a full 10-minute video, which is insanely cheap when you compare that against the old way of doing YouTube that I had to do when I was starting, where I had to literally hire every individual person from scratch and be paying anywhere from $100 to $300 per video. I actually have some channels I even pay way more than that, but on average I was paying a couple hundred dollars per video. And so the fact that we can do this now is absolutely incredible. All right. And now our video is done. So let's take a look at it right here. So it— made the images vertical due to the fact that the hotel front desk image that we uploaded was a vertical image. So it's got these black bars. Obviously, you would not want to upload this to YouTube. So we're going to go back to Claude and we're going to say, make a new version that's 1920 by 1080 for the talking headshots. Let me see if there's anything else that needs to be fixed right here. I also don't love the captions, so I'm actually going to ask it to remove those as well. Also remove the captions. They're really not that necessary. And we could spend time if we wanted to making them perfect by, you know, changing the font, changing how they are displayed, making them show up one word after another and so forth. But I'm just going to remove them right here because it kind of overlays on that hotel image right there too, which doesn't look the best. And then for the rest of the video, as you can see, we have our character right here. It's kind of transparent. I don't like that, like the actual image right here. So I'm going to actually screenshot this. And I'm going to tell it to Claude right here. Also, the attached image shows that it's kind of transparent when it's in the corner. Make it a bit bigger and not halfway transparent like that. And also say I don't like the black bars on the sides. There we go. Then we'll go ahead and enter. And what's cool about this is that we can make changes like this. So even if a video finishes, you still have those original clips generated. Higgsfield doesn't need to generate those individual ones again. But what Higgsfield will regenerate is the actual images of the character and where it's placed and everything else and fixing the black bars, generating new images without the black bars and so forth. And it'll just fix everything for us doing exactly what we asked. And now the video is done. So let's check it out. That fully booked room. Wow, this is, this is way better. Look at that front desk. Takes up the whole screen just like we wanted to. It's probably not fully booked. And we've got the lip sync right there. It's not perfect, but I think it's more than good enough for these types of videos. The minibar. If you wanted to, for example, save more credits as well, another thing you could do is to use better lip sync. Like, for example, there are software like Infinite Talk. I believe Infinite Talk is even built into Higgsfield. So if you wanted even better lip sync, you could do that. But let's go back over here. Did a soda you never touched. Sensors break all the time. I've worked behind a hotel front desk and there is a long list of things that property will never volunteer. All right. Now, looking at this right here, again, this isn't perfect. You can see some artifacts around the guy right there. Now, just as a proof of concept for this, what I'm going to say to Claude is I'm going to say, looks good, but please just generate now. 20 seconds of the video and fix the artifacts around the guy in the corner when he's in the corner. And I'm just doing 20 seconds just to show you an example for this here. Just have the plain PNG image with a transparent background for the guy in the corner. There we go. Just take a look what Claude is doing right here. It says the best way to fix the halo is to generate the host on a green screen background and then key out everything around him. So it's crazy that it's finding these solutions that otherwise I would have to research. Now, you could argue that this problem is only happening due to the fact that we're generating this with Claude. However, once we get a system working and once we get this working and it understands all the context of previous generations that you've done, then you'll have a system set up for you. You can customize this for exactly what you're looking for. You can say, for example, if you want the character appearing on the screen more, you can specify if you want it less, you can specify if you just want them in the corner of the video for the entire video. It's all up to what you're looking for, which can make it very confusing because you might wonder, what are the best practices for this? How can I prompt this correctly and how can I get the best practices and the best prompts to then put into Claude in order to do this correctly? And I'm going to show you how in just a second. But first, we're going to finish up rendering this video just so I can show you. And there we go. Now our video is done and it looks perfect. It's exactly what I was looking for. Let's take a look at it right here. So we have our lip sync in the beginning. Probably not fully booked. The minibar that detected a soda you never touched. Sensors break all the time. One thing I would specify is that if the character is being shown in the beginning, he shouldn't really only be shown for a second or two. So this would be a further prompt that we could specify. But look at this. It's exactly what we're looking for with the character in the corner, just like we asked. And it's all keyed out correctly. There's no weird artifacts around them. And there is a long list of things the property will never volunteer. And so now you know how to generate the right types of videos using Claude Code, but you can use any editor you'd like as well. I do highly recommend Higgsfield. If you do want to use this, you can get it at a discount at the link down in the description. And now how do you build a real business out of this? Because it's one thing to know how to generate the videos, but how do you specify the right things to Claude and make sure that it actually makes amazing videos and that you're building something with this where you can sell something from it and in turn make a lot of money potentially from this? I have tons of students who run completely faceless channels who make way more than full-time incomes from them because you don't just want to be running this for views. Views don't matter. What matters is how much you make and how much you actually keep from your channels. And so if you want to discover exactly how to do this the exact correct way and build this into a real business, then feel free to check out this video, which is a free course. It'll help you a ton and it'll outline how to make this a business. It's basically a free training. At the very least, if you don't watch it now, save it into your watch later. And make sure to go through it. Be well worth it. Hope you got a lot of value and I'll see you there.