I Created an AI Clone of Myself and the Results Were CREEPY

This is a daily upload channel and I've got a headband on. Welcome to the Parker Rex video. So, could I fully automate this YouTube channel to have four videos a day or just one video a day? Is that possible? What would it take? And would the quality be good enough such that I'd be comfortable cloning myself? Let's get into that today because I'm interested in it. There's technology that's definitely capable, but can I tweak it the way that I want to? So, this is Daily Upload channel. My name's Parker Rex and we're going to cover all things clones. We will answer questions if we get them. Lately, it's just been nice things that people are saying. So, thank you for the nice comments. Josh asks, "Does augment truncate your code before sending to LLM like cursor does?" As far as I know, no. If you read their articles, they have a great blog, by the way, and they cover how they do this. They have what I like to think of it as kind of like a state machine for your codebase and that is always being updated. So their bread and butter is actually large code bases and that's why I find it to be the best. I still want to stay up to date with what's going on in cursor. So I'll be making some videos about that this week on my main channel. Dan says pantic is great use for what I was referencing in a video where I want to make a T3 stack for fast API with next. And Dan again says deep research agent party. Now, Dan, if you could be a lad and just shoot me a Discord message with any of the prompts that you're using because I'd love to know. Mike says, "So helpful getting to see your process, the research, and think live. Never realize that watching someone else learn is a great way to learn." Yeah, I also recommend two other channels that are really good that I enjoy watching that I learned a lot from is everybody probably knows Tio Tio T3 is what I meant to say but Theo and he is I think this if you just type in some iteration of those you'll find him webdev Cody phenomenal these are both YouTubers and then there's a guy named Shay and I think there's like a four in it and he's building user jot and I've learned a lot about doing different ways of deploying from him. He has a self-host blog. He also just has really good Twitter. And then finally, Dax is another one. So, he's these are the Twitter guys. X.com. And then for YouTube, these are my two favorites. So, let's get into the beef of the video. So my question is what does a templatized Parker Rex daily video look like such that I would increase the quality so much on the production by building in the Lego blocks that maybe the AI of it would be at a high enough quality that I'd feel comfortable doing it. So, well, I think there'd be some sort of intro. I want them to stand out so I could do a front flip on the ground. Just wanted to share that. Really, that's the excuse. No, but I thought of it where you could have some sort of crazy scene at the beginning where you would have if you storyboarded this out, if it's just 4 seconds long and you had just a couple of frames to guide on it, that you would have the subject aka me cuz you're trying to grab people in the hook, but you could have the subject kind of in the distance. So, it'd be me here and then I'm running towards the camera. So I get larger and you know you do the thing you do something flashy and as soon as I then hit the ground the scene shifts up and this is when the AI stuff kicks and you have me falling through some sort of roof and then into this scene that you see here. Now, I don't think that V3 can do this, so I probably won't do it, but some sort of intro and then I found a good example of this that I liked. So, with GitHub, they have an intro and they just have a jingle and I'll talk. So, instead of wasting my money, I'll just come in, experiment on GitHub's money, and then figure out which model is going to be what I want to work with on my app. Yeah, exactly. If we give you stuff for free, then why would you pay for it? Hey developers. So that little beep beep beep beep beep. Now it seems silly, but there's examples of tone that are across the most popular companies that we see that we buy stuff from where they'll use a major key and it will be an example of McDonald's. I'm loving it. It was ba ba ba ba 1 2 3 4 5 and then sprint or metro PCS was likeun dun dun dun dun. So something that's just a little tone that gets you to remember what it is and stand out. And so I'd probably make something in Ableton Live. Now I used to do music production if you didn't know. But you pick some sort of sound. And then I could also have different sounds towards the end of the video or during scene changes. So the only two scenes that I have are this and this. But if we do those, we can time them to a different sort of sound. So if I came in here and you when you music produce you basically just do in these 4 8 16 32 blocks and you can start making sounds and just press the key on my keyboard. So major key would be 1 4 and 7. So if we go here and let's do it's four five six seven something like that. And then you can play around with actual sounds. But that's one aspect of what I'd be thinking about is like how could I make this sound more interesting and [Music] memorable and then it would cut. So I'll continue to play around with these ideas. Or you could even do like a voice recording of me talking and then I could have it go to something else. For instance, let's put here. We'll do a recording Rex video. Cool. So, let's crank this up. And that's so funny. I haven't used this in forever. Welcome to a Parker Rex video. We could throw a vocoder on this. Crazy. Welcome to a Parker Rex video. And then I think with the actual AI bit, you need some sort of data pipeline. So, we've covered the sound, maybe some sort of intro. or we can use Ableton Live to do some sort of arpeggiator or something interesting. But what I'm thinking then is you essentially have a pipeline that's always looking for stuff that's interesting that can cover the news. So you could do a GCP setup and in here we can go and find a good example of this. So let's go to GitHub generative AI marketing repository. Okay, here this is exactly what I was looking for. So you do something like this where you use gel news scraping and you can feed in the vector AI search for different topics. So thinking adding different news sources that I like and things that I look for. So that's typically like my GitHub that's typically my GitHub feed and then also X and I could combine that with something like this bright data website. So, this allows you to do proxies and scraping. And I could go and target Twitter, target the actual lists that I have. So, I could make the list public and then pass that in. And then you'd go and you'd have all that information there. And you'd set some sort of timer, cron, whatever it is. you'd have a job that is going to be cued for each day so that each video could go live in at 8 a.m., let's say. So, I think that's a pretty interesting way to do that. And then last is the most interesting one. So, we're going to test this out. This is Dcript, and I want to test their new AI stuff. So, let's get that pulled up. And I know that they have a voice thing you can do. So, let me come into here. I just want to try. Can I make something out of a video or out of Can I type something? Let's see. I need to write a script. Let me see. If I come into here, I can do AI video maker. I can upload a file or paste in a script. Okay. Generate generate a video from a prompt. Okay. This video is going to cover GitHub models API and what it's capable of. It just recently launched and it allows you to see the performance of your GitHub models that are being used for your particular application. What this means as a developer is instead of building all your own custom tooling and seeing how your prompts are performing, it will do that for you inside of GitHub. It's currently in preview. It's been out for several months, but they continue to add more and more features and I think this will be great for everybody's workflow and it went away. So, let's open up Whisper Flow and let's grab this. Let's go back into let's create AI video. Paste in this script. Continue. And now I should be able to make my own avatar out of a photo. So let's go into here. I need to make a new one. No avatar. How can I make my own? Let's see how this works. Sure. Car. Why not? Now it's making the title, dividing it into scenes, choosing my narrator. So the next obvious step would be make my avatar. Now I watched a video on how to do this before. And they basically just have you upload a photo, but I'd much rather it be trained on the videos that I record because it'll be more accurate. And then for a voice, same thing where they give you a prompt and it says, "Hey, here's what you need to say." So, it's kind of like the quick brown fox jumps over the lazy dog. That typing test thing that covers all the words, all the letters in alphabet. So, they have a set of sentences that you read out, but let's see how this does. So, it's generating my avatar. That'll probably be the most time consuming thing. So, let's pop open a new window. We're going to test out create with AI speaker. How can I do this? I want to make one for me. Make a new speaker. Kyle, let's test. I want Dcript to create an artificial version of my voice that I can use to create speech that sounds like me. I'm training my voice by reading the following statement. Imagine a big blue ball spinning in space. That's our Earth. On it, there are tall mountains, deep oceans, and huge forests with animals. In cities, people talk, play, and work. We all have different voices. Some people speak softly while others loud. Every day, we tell stories, ask questions, and share jokes. They can have my speech. Now, the privacy nuts would say, "You're crazy for doing this." But the reality is, if you post content on the internet, it's already trainable. Someone else can do it. I'd rather do it myself and not let someone else. So, while this is generating, I will pause the video and come back. So, the trained voice is complete. So, we're going to go and choose that as a speaker. And now I get the chance to upload a screenshot of my face. So, let's grab one of those and go into raw OBS. See if we can't find Okay, there we go. Straight on seems fine enough. And we'll assign that avatar. Looking freaky. I probably need something with my teeth, I would imagine. Yeah, that's probably better. Can only do one. Seems strange. Let's grab that. Okay. Assign avatar. And now I can start writing. So I'm going to grab a script from a previous one. And let's go into here. We'll just grab portion of it. Save this. Come into here. Okay. Testing AI avatar test done. Invalid characters. Okay. Generating the speech. And cool. We'll come back in a sec. So, it generated the speech. Let's give that a listen and then I'll hit generate avatar. You got to build your own stuff. That's how you get better. When it's painful, you're learning. Okay. So, let's talk about how to do that. This is a daily upload channel and I've got a headband on. So, I'm a ninja now and I want to talk about the template that I'm building because that is so obviously AI. I would never publish that. Let's do a quick comparison so we can see how bad that is and why you'd have to build your own tooling. This is why I'm building my own tooling for this because I think with the right stuff you can do this, but I'd need to feed so much video so that it have the inflections and the lows and the highs and all those pauses, slower words, faster words, stuff like that. So, let's go back into here. And on the left is the AI one. And on the right is me. And I'm curious if I can generate the avatar while it's going. But let's play real Parker. You got to build your own stuff. That's how you get better. When it's painful, you're learning. Okay. So, let's talk about how to do that. Even that word lets. So, let's let's versus let's let's talk about how to I need to make that a little bit louder so you can hear studio sound. You can't even do studio sound. Okay. Can I? Weird. Let's generate the avatar. You guys heard it. It sound awful. Got eight minutes left. Okay. About the Okay. Okay. And here's the generated avatar that they wanted me to use. So, let's check this out and see just how dog water it is. Hello everyone and welcome to this video about the GitHub models API and its exciting capabilities. Now, recently launched, this API is designed to help developers like you track and analyze the performance of your GitHub models that are crucial to your applications. Instead of investing time and resources into building custom tools to monitor how your prompts are performing, GitHub now provides a seamless solution directly within their platform. This feature is currently in preview and while it's been available for a few months, the developers at GitHub continue to enhance it with even more features to improve your workflow. As developers, we constantly seek efficient and effective ways to streamline our work. What I do think is interesting is how they do this like scene detection stuff. So like the way that that animated in and whatnot. So it's doing a lot of transition work, which is nice. But how would I then take it and make it me? So we have this. And how do we avatar Cedric? Hi Cedric. You're so creepy. No, I hate it. No, no, no, no, no. And I'm really going to hate when it gives me a fake me. It's just not it. But we'll see. Okay, so it's completed. Let's have the big moment here. Oh boy. You got to build your own stuff. That's how Hang on. That's crazy. How do I crank this up? They don't not make it very easy to edit. Let's see. Layer audio. audio effects limiter negative3 and let's throw a compressor on it so we can get it louder heavy. You got to build your own. You got to build your own stuff. That's how you get better. When it's painful, you're learning. Okay. So, let's talk about how to do that. This is a daily upload channel and I've got a headband on. So, I'm a ninja now and I want to talk about the template that I'm building because I think we as developers, especially on the newer side, are always looking for a template or we're always looking to save time really regardless. Oh, that is so creepy. Yeah, I don't like that at all. Worth a try. Definitely would not use this at all. I I know it's going to get better, but it's not there yet. If you guys learned anything in this video, make sure you like the video, make sure you subscribe to the channel, and you can check out VI if you'd like to learn more. All right, see you