I Let GPT-6 Astra Run My Faceless YouTube Channel For 60 Minutes and THIS Happened

ChatGPT Astra 6 just came out and it's OpenAI's latest model that a lot of people are saying is even better than the Claude Fable models. Now, I did a video on my channel here a few months ago where I let Fable 5 create an entire YouTube channel and make videos for it, and the results were pretty amazing. And now we're at a point where I'm basically using AI in almost every single part of my YouTube business with the exception of ideation and refining scripts. But Fable 5 and the newest Fable 5.1 on Claude that came out are so incredible. And they've been helping me so much that when I heard about Astra 6 potentially being even better than both of those, I had to give it a try myself. And I had to see if it lived up to the hype. And in this video, I'm going to be using Astra 6 to run my faceless YouTube channel for an entire hour. Now, if you want to follow along with me in doing this, and you want to run a YouTube channel using AI yourself, you will need a few tools, the first of which that I'm using is ChatGPT itself, you have to be on one of their paid plans, and you can access it by going to ChatGPT and then clicking the work icon right there. And then you can select Astra 6. And from here, you can set what is called the thinking power. Now, the higher you put this, the faster you'll run out of your weekly usage. So keep that in mind. But for the purpose of this test, I'm gonna be setting it to both a mixture of max and high, depending on the tasks that I'm doing. And the next tool you're gonna need if you wanna follow along with this is a tool that will allow you to actually generate videos. ChatGPT by itself does not do that. In fact, when they tried with Sora, they lost so much money that they closed that down. But for this video, I'm going to be using the sponsor of the video, which is Higgsfield, to generate the videos and to be used within ChatGPT. And in order to connect the tools, you go into ChatGPT, you click Plugins, go to Connect Plugins, Browse All, and then from here you can go ahead and search Higgsfield. This will then prompt you to connect your Higgsfield account and sign up for one if you're not already. You can use my link down in the description if you want to get a really good deal on it, by the way. But now, just like that, ChatGPT has access to all of the newest AI models, including Seedance 2.5 and all of the other models that allow you to generate incredible quality videos. Now, before we go ahead and give GPT-6 Astra full reins of my entire faceless YouTube channel and see what it does, let's first start off by testing the scripting capabilities of the tool. Now for most of the scripts that I personally currently write on my YouTube channels, I use Claude and I use it through an interface that I built called TubeMagic. In my opinion, the script is the most important part of a video. In fact, you could have subpar visuals, but if you have a really good script that gives a ton of value more than any other video within the niche, it can make up for a lot and it will get you a very high audience retention if you structure it correctly and therefore a lot of views. But I want to test the scripting capabilities of this new model just to see how good it is. But before I do, the channel that I'm going to be letting ChatGPT run for an entire hour and do all this on is my own faceless YouTube channel called Stickman Matt. Now, I don't share the names of all of my faceless channels. I don't want to pollute them with people who aren't interested in the content. But Stickman Matt actually teaches how to grow on YouTube and how to make money on YouTube, which is the exact audience of this channel. So if you guys want to go check that out, you'll get a lot of value from the channel as well, and it'll be a win-win for both of us. So can GPT-6 Astra run this channel just as good as a human can? Let's take a look and let's start off with doing the script. So I went into ChatGPT right here. I set it to maximum thinking power and I put in this prompt. Please write me a 5-minute YouTube script for a video called Do This and the YouTube Algorithm Will Actually Push You in 2026. This will be for my Stickman Matt channel, which is me but in stickman form. Attached is training data of my teachings, and I attached it with this file, which is basically a compilation of all of my different YouTube videos that I've had. Now, if you don't have this, don't worry, you can easily get this from other people's videos as well, but you'll want to rearrange it and remix it to be your own, which is what AI can help you do. And then I asked it to just give me the text, nothing else, don't mention cuts or anything else, just the script text only, so then I can have you make a full video from it. And it gave me this text right here in a downloadable file that I can download to my computer. And it does get straight to the point. It says, if you want YouTube to push your videos in 2026, start with a topic people already want, give them a clear reason to click, and then deliver on that promise. Then use what worked to plan your next upload. Here's how I do this in 5 steps. That's pretty good. It's straight to the point. It might not be the hook that I personally would write, which you think it would be a little bit more in my style due to the fact that I've uploaded so much training data right here. However, it's not bad either. I would see this getting a pretty good audience retention. Now if I compare that to my own YouTube tool, which is TubeMagic right here, which we're going to be building in Astra 6 as well as the latest Claude Fable models and everything else into right here, and we look at this intro: most creators are doing everything wrong on YouTube right now and they have no idea. They're uploading consistently, following all the advice from 2024, And then wondering why their channels are completely dead. Here's the truth that nobody wants to tell you: the YouTube algorithm in 2026 is fundamentally different from what it was even 12 months ago, and if you don't adapt, you will get buried. That's really good. In fact, that's more in tune with how I would actually write this script, and that's due to the fact that we build all of these frameworks within TubeMagic. Now, you don't have to use TubeMagic for this; you can use GPT-6 Astra. But I just want to tell you, whenever you're writing scripts with AI or even using it just to make a first draft, which then you can later edit on, make sure that you. Train it with a lot of data. Don't just tell it, make a script on this and trust that it's gonna be good. Give it transcripts of tons of other videos out there within your niche that have done really good. And in the case of my tool, I've trained it on all the best viral hooks and intro frameworks, value loop structure, strategic ordering, and all of these things, which will optimize the script for the algorithm and more importantly for people who will watch more, getting you more audience retention and therefore more views. But in the spirit of this video, I'm gonna use the GPT-6 Astra script just because I wanna see what this tool can make if I just let it do absolutely everything. So I'm going to go ahead and download this TXT file right here. And so then I made a new chat and I told it, please make this guy a saved character within Higgsfield that I can reuse and call him Stickman Matt. This is the stickman that I actually use from my channel. And I generated this image right here using GPT Image 2 within here. And now we're actually up to GPT Image 2.5, which is even better. And that's also included within ChatGPT. So I've uploaded these 2 images and I asked it to create a character within Higgsfield. This is going to allow me to then reuse that character, which allows me to build brand IP. So even if millions of other people, you know, make copycat videos of this and everything, I'm still building brand IP and a connection with viewers with my character doing this method right here. You don't want to just make commoditized videos, especially in the age of AI. You want to make sure that you're using a character that you can reuse on every single video on your channel. And that's what I did. So then it added this to my elements on Higgsfield and it saved it. And then I uploaded an audio file of me talking. This was actually from ElevenLabs. I had it just say some things in my voice. You can even use my voice as well. It's on ElevenLabs for you to use. But I asked it to then create a voice within Higgsfield called Matt and use the attached voice audio to train it. And then it said it can't do that directly within here. So I simply logged into Higgsfield and I uploaded it under the audio tab and I created a custom voice, and then I can do everything else within this ChatGPT interface. And now in order to make the best video possible, what I did next was I downloaded one of my videos from Stickman Matt and I uploaded it to ChatGPT, and I told it, analyze this video and see how it's structured and how many cuts there are and make a detailed analysis so you can then go ahead and make videos that look and sound exactly like this when I ask you to. And it went in here and it saw like how long is the average visual segment, which is 5.9 seconds, and the narration and everything else and the visual balance and all these things, which I had to pay somebody to do all this in the past. But now it's remembering all of these things. So then I can use this model to do what in the past would've taken me days of work and a whole team doing this. So then it created this style guide right here, which is perfect. And from there then I told it that I'm gonna have you make a 5-minute video following that using Seedance 2.5 720p, but don't make it yet. I want you to estimate how many credits this will cost so that I can buy them on Higgsfield. And it told me exactly how many credits that I will need. And then I said, according to the average Higgsfield plan, how much would a 5-minute video cost? And it gave a dollar equivalent right here of somewhere around $100, $94 right there. So Keep in mind, that's how much you can expect to pay per really good quality video. Now, Seedance 2.5 is the most expensive model and it's the best. You can easily use much cheaper models to get basically the same look. For example, Seedance 2.0 is pretty much just as good for videos like this, which would be a lot cheaper. You'd probably be looking at half the price of this. And there's even cheaper models than that that can do incredibly good, like Kling 3.0. So you can realistically get these videos made for anywhere from $30 to $50 if you use some cheaper models. Models, but I want to use the best. I just want to see what it can do here. So I told it that, and this is still much cheaper than paying the team because with the team I was paying several hundred dollars per video. And we're going to compare at the end of this the video that I paid hundreds of dollars for and this video which was less than $100, and we're going to see which one was better. And so from there then I uploaded my YouTube script and I told it let's make the actual full video. The script is attached. Please use the Matt voice in Higgsfield and use cDance 2.5 720p for the whole video, and using the research you got earlier in this chat for how to structure everything and keep it engaging. No music, please. Give me the finished full video back. Of course, you can use music as well if you want. Just tell it to use royalty-free music, but I'd prefer to either do that later or do no music at all. I don't do music on a lot of my videos, especially if they're educational in nature like this. And so then it went through and it generated all of the different clips of our stickman character in his office talking. And then it even generated visuals outside of that. It stitched it all together and it put the entire video together with the proper narration in the proper voice into a full video that was 5 minutes long. Now this entire process took around an hour and 22 minutes for a 5-minute video. Now that it's trained on this and everything, I do foresee it being faster in the future. If I were to do another one of these, maybe it'll only be 30 minutes, but still only an hour of work. Whereas in the past, I would have had to pay an entire team and then wait, you know, 2 to 3, 4, 5 days later to hopefully get the video back. I get it back in an hour that I can instantly tell it to make improvements on and so forth. But I want to see what it can do in one shot. And so let's take a look at this right here. Now the video finished, but it didn't actually give me the video. So I asked it, where's the finished video? I can't see it anywhere. You said it's finished. And then it gave the video right here. My internet actually cut out as I was doing this. So that might have been the problem. And it might have even been much faster than this, to be honest. But it gave me the full video and it was exactly 5 minutes and 6 seconds, which is crazy. And it gives me a link to download it right here. And here is the video.

If you want YouTube to push your videos in 2026—

I just want to point out that that lip syncing and mouth movement is absolutely perfect. So if you use a cheaper model like Klink 3.0, the mouth is not going to be nearly as good. It's just going to be likely an opening and closing mouth and the emotions and everything might not match up. You're going to see here in a second Good the emotions are, but I just want to point that out. Let's continue.

If you want YouTube to push your videos in 2026, start with a topic people already want, give them a clear reason to click, and deliver on that promise. Then use what worked to plan your next upload. Here's how I teach this in 5 steps. Number 1, find proven demand. When I started YouTube, I struggled for months.

So you can see that emotion right there when he's talking about struggling, his eyebrows kind of go down, that's perfect.

Then I made a better version of a video idea that already had millions of views. My version got 100,000 views in less than a week. The change started with the idea.

And this is actually accurate, by the way, as well. This is actually pulling from my story, so that's correct.

Pick one specific audience. For example, beginners building a faceless YouTube channel. Every topic should solve a problem that audience actually has. Find channels serving those people and look for videos that outperform their usual uploads. Say a channel normally gets 2,000 views, but one recent—

I love how it even says illustrative example right here. So just to give a little bit of a disclaimer that, you know, not every single video or every single channel will be like this. I really like that.

Video gets 40,000.

This is great.

That's worth investigating. Compare videos that have had similar time to collect views. Study the topic, the title.

Wow, I love— he's even showing his fingers like, study the topic, the title. That's awesome.

And the problem it solves.

Wow.

Look for that demand across several channels. Number 2, make your version worth choosing. Watch 3 strong videos on your topic. Write down—

Yeah, this is fantastic. The only thing that I would say regarding this is I would put some more cuts in it and make it a little bit more engaging. You'll notice with this video right here, when I send it to my editors, they make sure to, you know, zoom in on my face, zoom out, keep it a little bit more engaging. And if I was going to upload this video on the actual channel, I would want the camera to push in closer to him at times and so forth. And This would be something that I can easily tell ChatGPT to do. Now, you don't want to generate a whole video and then tell it to refine it later because then you're spending another $100. However, what you can do is generate a sample video. This is what you realistically should do if you're going to be using this, which I personally am. So this is what I'm going to be doing. I'm going to make some very short videos like 20-30 seconds long and do a series of tests just refining it and making it more and more engaging, telling it to do faster cuts every few seconds. Telling it to zoom in, maybe use some sound effects as well, and get a much better video. And then tell it, okay, now make a full 5-minute video based off of that. But overall, this is a really good video, and these graphics are fantastic. I absolutely love them. It even shows the 33 videos. I wonder if this is actually accurate. 3, 6, 9, 10, 11, 12, 13, 14, 15. Yes, yes, it is. That's actually 33 videos. That's awesome. The expressions and everything are perfect. Graphics are perfect. I love it. It's a perfect video. Now, of course, if you wanted this styled in a certain way, you can tell ChatGPT and it can do that. But we just made an avatar video with an actual avatar with a face on it, an entire video within an hour. And now if we compare this to the video that I made with a team with using AI, where basically I gave my team AI tools, but they still did all the finished editing and everything, and we watch what they made. Let's take a look.

Growth video tells you the same things. Post consistently.

So these are all AI graphics that were generated using different models like Seedream 4.0 and Nano Banana 2, but they were edited together. They put sound effects on it.

Your thumbnails, write better titles. And if you've been doing all of that for 6 months and your channel is still stuck—

But as you can see, the lip syncing is a lot worse on this one, and that's because they were using a cheaper model likely to do that. There's some more varying in the locations that the stickman's in and everything. And maybe you could argue it's a bit more engaging, but overall, this is getting really close. And I think with a little bit of tweaking, I can make this video just as good. But honestly, as a viewer, personally, I would rather watch this video right here, the one that we made with GPT-6 Astra, because it's more straight to the point and it gives really good graphics that actually illustrate what's going on as opposed to just having, you know, tons of different fancy backgrounds and everything. But this one does give some graphics here as well, which are quite good. But overall, we're at a point now where you can use AI to basically make any type of video that you want, but it just comes down to how you use the tools. Now, the best tools right now are all on Higgsfield, and if you want to check them out, click the link down in the description of this video, and you can use them on any AI model, whether it's Claude or GPT-6 Astra or directly within their interface. They're an integral part of so many of my different YouTube channels. But now that we're at a point where anybody can do this and anybody can just make any type of videos, you might be worried like, are videos just gonna become commoditized to the point where it's gonna get to a point where we click a button and get a super high-quality video? And the answer is yes, it will. However, you can make your channels stand out from that by having a character as a face of the channel, building a connection with that character, and selling something from your channel that helps people more in depth in a more step-by-step way that leads them through chronologically whatever your channel is teaching. And the way you'll do that is by running an educational channel. You can, of course, still do entertainment channels. For example, there's tons of channels succeeding doing like AI short films and everything else. Those can do really good. You can get a lot of views and make a lot of money from ad revenue. But the majority of the money to be made on YouTube is with educational content. So if you have any skill that you can teach other people or anything you're interested in learning yourself, make a YouTube channel around it, make it educational, sell something from your channel and make a ton of money. And if you want to see more in depth exactly how I do this process on tons of different channels, and I help channels build real full-time incomes all the time, you can check out the video next to me right here, which is basically my completely free course for doing this. Check it out. My name is Matt Parr. Hope you get a ton of value from it, and I will see you there.