Transcription
I've got some marvelous, remarkable, groundbreaking developments in AI to bring you this week. We're talking about AI musicians that are taking over Spotify, a brand new video model that's blurring the lines between real and generated, as well as an incredible new text-to-3D software that's bringing the believability of AI games into reality.
First up, AI music is officially going mainstream. There's a band on Spotify right now, Velvet Sundown, which are racking up over 750,000 monthly listeners. But here's the intriguing thing: It appears to be entirely AI generated. There's no label; there's been no tour, and there are no real people. Now, their team denies it, but fans, devs, and even streaming platforms are picking apart the evidence. Let's take a closer look ourselves, because this might be the future of music, and nobody even told us it had begun.
So, this is the Velvet Sundown, and you can see now they're up to 858,000 monthly listeners, and their image on their profile is obviously AI generated. I think it's extraordinarily clear to me. There is something about the quality of the skin tones that you can recognize as AI generated. Now, even their bio on Spotify has some hallmarks of AI generation, notably this EM dash. Now, one of the important elements here is, of course, that on Spotify, the Velvet Sundown are given a verified artist tick, which draws into question whether we can trust Spotify to be telling us when it is AI-generated content.
Now, just how this band managed to rack up an incredible amount of plays in such a short amount of time without any online following or any marketing remains to be seen. However, the news attention that it has garnered just in the last week has shown its streams go from 300,000 to more than 800,000. So, this could be a simple marketing ploy, and there may even be a plant inside of Spotify who is promoting this via Spotify playlists that is allowing it to garner a viewership, or a listenership, you might say.
Now, if you're interested to see how they might have done this, it's actually incredibly easy. It's most likely they use a platform like Sununo, which allows you to put in a simple text prompt and get out an AI-generated piece of music. So I will put in a simple prompt like "an indie psych rock band." And in just a couple of seconds, you get two song options. And here is my first offering. "Hey man, I'm Silk Sunrise. I've just been generating some music with my AI-powered drum." [Music] "Lights bleed into the dark blur of the street," [Music] "faces melting into the pavement heat."
Now, let's talk numbers. We're up to 1 million monthly listeners for The Velvet Sundown. And this might sound impressive, but how much money is that actually making? Well, depending on the payout rate, that's somewhere between about three and $4,000 a month. Now, it's not life-changing money, but that is an incredibly useful amount of revenue to be generating from an AI asset, especially when you factor in you could be creating other revenue streams related to this. So, let's actually take a listen to one of their songs. [Music] [Music] "Dust on the wind, boots on the ground." Now, it's particularly good. And what's impressive is how well it maintains consistency of the genre, the style, and the vocals between different tracks. There is a feeling of cohesion across the whole album.
But this is not the only AI artist starting to show up in the media. Timberland, a record producer and rapper, has recently released another AI artist called Tata, and it is an attractive and photogenic Asian lady. And you can see that this is starting to become more and more prevalent. And I think it's an incredible opportunity, especially if you're leveraging all of the latest-gen AI tools together. And I do have a course on creating your own AI influencer online. And we're going to be adding a module on how to create your own music for your own AI influencer, too. You can check that out in the link in the description.
Now, another interesting part about this is it's almost certainly AI generated. And there's a very good video from Rick Beto where he breaks down the track from the Velvet Sundown using Logic Pro on his Mac. And this separates out the different elements of the music. And when doing so, it becomes obvious that this has been AI generated because there are huge amounts of artifacts, and the software highly struggles to do this for this track. I also think if we listen to it that, for me—I'll be interested to hear what you say in the comments—but for me, this certainly feels like a synthetic or almost a robotic nature, especially to the vocals.
Now, music is not the only medium where we're seeing massive developments in AI right now. Take a look at this. This open-source AI model from Han Yuan is allowing you to input simple text prompts and get out detailed 3D assets. Now, it also works from taking a simple image and turning it into a 3D model. Now, this opens the way for generating complex 3D environments simply with text prompts. And this is an arena where we're going to see huge developments in Gen AI right now because we're being able to create more complex, more diverse, multimodal pieces of media, which involves creating 3D environments that are also interactive.
And that brings me on to the next project, and that is VM, which is a consistent interactive video scene generation with surfel indexed view memory. Well, what does that mean? It means that the video manages to remember exactly the layouts and environments that it is generating, and this allows it to return to places in the video. So why is this important? Well, most AI video tools forget what happened earlier in a scene or struggle to keep things consistent when the camera moves. So if it pans one way and then pans back, the reality changes. This new plug-and-play module fixes that by remembering what the scene looked like. It anchors it to 3D points, and that way it can generate new views more accurately and with less computing power. That means smoother, more realistic, and more coherent AI-generated experiences, especially for long scenes.
So let's have a look at this in practice. Here we have three different images. The first is the input image. The second is what happens when you create a video from this scene. And you can see, as it looks forward and then goes back, the environment changes—that this is now an outdoor view where previously it was another passageway. Whereas with this new module employed, you can see that it maintains consistency. You can show this in another example here with an office situation. And as you can see in the first example, as the camera pans back around, you can see that it doesn't make any sense. Whereas with VM, it maintains the adherence to the 3D environment. Now it does this by maintaining a memory index of the key points from the previous scenes.
However, for 3D environments, we are still needing higher-quality video generators. And this model is one that's blowing away people on the internet. You may have seen some of these viral cat-diving videos on the internet in the past week, and these have been created by Halo 2, which is today's video sponsor. Halo Video 2 is now ranked number two on the AI text video leaderboard. That's just behind Sora and ahead of Google V3 and Cling 2. Now, Halo Video is also sometimes known as Miniax. And what's particularly interesting about this is this is a blind test. So users are shown two videos at the same time and a prompt and asked which one do they prefer. And with this model, it's outperforming some of the best AI video models out there.
Now, interestingly, it also follows prompts with an 85% success rate, even on complex or nuanced instructions. So that means if you have complex movements that you specifically need, it does an excellent job at adhering to those. Now, the developments of Miniax mean that motion is more realistic. Just take a look at these acrobatic actions and challenging physical movements. There is very little distortion at all with this. Now, another feature that's leading the way inside of Halo is character consistency, which holds well across a number of angles and scenes. Visual quality in Halo is strong. I particularly admire the cinematic lighting, stable composition, and the clean depth of field. Halo supports 10 seconds at 768p and 6 seconds at 1080p.
Now, if you're looking at price, the benefits of Halo is that, per second, amongst the other top-tier models, it's cheaper than VO, Dubau, or Cling. Now, this is incredibly useful, and it really puts Halo at the top of the pile. It's one of my favorite AI video models, and the best thing that I love about it is that they're offering new users 500 credits for free. If you sign up today, Halo's focus is clear: high performance, low cost, and open access. Now, you can click the link in the description below to try Halo AI for yourself today. And I would like to say a big thanks to Halo for sponsoring this segment of the video.
Now, another viral type of content that has been exploding recently is ASMR AI video. And some of these videos have been racking up tens of millions of views across X and Twitter. Let's take a look at some of these. This is one of the biggest trends, which is glass fruits being cut. And what's particularly interesting about this is just how well the physics is maintained for these quite complex situations. And you can see that the type of material is playing a large role in how the fruit is dissected.
Now, you can create these for yourself. And here is a prompt that I've been playing with that creates extremely excellent ASMR-style videos. Now, all you have to do is change the parameters inside of the prompt. So, you can pick your own type of fruit or other object. So, I've generated this as a glass burger inside of Halo using their brand new model. So, here is my beautiful glass burger. And as you can see, it's wonderfully sliced.
Now, what's really fun is you can actually add audio to this using Eleven Labs. All you have to do is come to Eleven Labs, upload your video. It will analyze the video, and it will generate a custom soundtrack for this. Now, this is a wonderful way to add a real sense of immersion into your ASMR videos. You get a whole host of different sounds. "Oh, that one's good. I felt it." And it's as simple as that.
Now, Eleven Labs also have another very interesting release that has been doing the rounds on social media. So, let's take a look at how this will potentially work. "Hi, Eleven. Good morning, Jay. Uh, when's my first meeting? It's in 5 minutes. Who's he with? It's with your boss, Jay, for your 2025 annual evaluation performance review. Oh, great. Uh, email them and tell them I'll be 5 minutes late. Email delivered. Oh, send a heart emoji to whatever my mom sent me last night. Heart emoji delivered. Oh, yeah. Can you order the usual for breakfast, please? The vegan big boy from Tony's. Oh, yeah. I could do with that today. Anything else I can help with, Jay? Actually, yeah. Pull up my boss's socials. Okay, I'm ready when you are. Didn't he uh get a Bengal cat or something? On the 29th, your boss posted that he adopted a Labrador. Okay, great. Thanks, Eleven. That's perfect. Try 11 today, your AI personal assistant from 11 Labs. Hey, boss. How's the new dog?"
Now, why this is incredible is because you can connect it to your existing apps, and it can pull in relevant information to your life. And this allows us to have a much more powerful AI assistant. Now, to use this, you simply have to come to Eleven Labs, sign in, and go ahead and build your own personal assistant. Now, they've got a lovely interface, and you get to start off by giving it a name. I think I'll call mine Noodle Face. Then, you can choose a voice. It suggests a few options. "I would like by repeating what students say. So essentially, well, what I'm all right, are we cracking code, solving mysteries? Let's go with Lucy or Noodle Face." Then you can go ahead and create your own Noodle Face.
Now, the next step is to start adding in your own integrations. Now, it talks about MCP server integrations, and this allows you to connect other services into the system. However, what's important about this is to recognize that it may involve the sharing of data with third-party services. And of course, this is something we must be aware of with all of the information that we are now sharing with these AI platforms. I mean, I certainly tell things to my AI that I would probably not tell to anybody else. So, you have the option to add your calendar. You can add Notion, Perplexity, and Slack. And this allows you to start adding an agentic style of interaction with your AI.
So, I've just added my decal, and I'll show you how this works. So, you can go with "start a call." "Hello, Samson. Feeling productive? Let me know how I can help." "Great. I'd also like you to make an appointment this afternoon at 4:30 p.m. and invite the following Gmail account." "Got it, Samson. I can definitely set up that appointment for you. And what should the title of this appointment be? And how long should it last?" "A date with my noodle face." "That's a lovely title, Samson. A date with my noodle face." "It is. So, starting at 4:30 p.m. today, how long should our date last?" "I would like it to be 10 minutes with us." So, right here, you can notice I have to approve these manually, but you can turn that off in settings.
Now, what's remarkable is that if we go into my calendar, these now exist inside of my calendar. And you can see it's already invited the email that I asked, though, it did spell it wrong. So, that's why you have to make sure that you pay attention to exactly what's going on. And generally, it does require a final layer of approval. However, this is quite remarkable, and the potential for this for automating and allowing you to do a lot more conversationally, I think is so exciting because, for me personally, I find it quite challenging to spend so much time indoors looking at a screen. And what I've been trying to do in my life is to use AI to allow me to do more work without looking at screens. And it's so exciting that you can actually be incredibly productive simply by going for a walk and talking to your AI and getting things done. And I actually find this to be a much more inspiring and enlightening way to interact with technology.
Now, another intriguing development in the world of AI is that robot footballers in China have staged the first of a series of three-on-three matches where the robots are operated autonomously. So these are autonomous robots playing soccer or football, and the incredible thing about this is that they are working completely on their own as a team. Now, there is some comical and humorous falling over of the robots, but the impressive thing is is that they were—they are able to coordinate as a team to achieve a goal. Now, this is one of my favorite moments because they have to bring in a stretcher for the robot and pull them off to the side. Now, for some reason, I find it absolutely fascinating to watch these robots play human sports. Especially because there is a little bit of glee knowing that, at least for the moment, that humans are better at playing soccer than the robots that are playing. But what it does pave the way for is an exciting emergence in the years to come of combining our latest AI technologies into that of robotic bodies.
Now, an interesting good news story in the world of AI: In 2024, 21-year-old Roy Lee was expelled from Colombia for developing an AI tool that could ace interviews, exams, and meetings. So, what it would do would be that he would be having an interview, and he would be getting AI assistance on the side for these interviews. So, he was asked difficult, challenging questions or tests; he could get AI support instantly. Now, a few months later, he rebranded the tool as Cluey, an AI assistant that helps users crush tests and deliver perfect pitches. Now, in June 2025, he has secured a $15 million investment led by A16Z, pushing the valuation of his project to $120 million. And for me, this shows the opportunity for what we can all start to do with AI. We're living in a wonderful era of possibility. And if you are interested in diving deeper into creating real projects with AI, I invite you to check out my courses in the description below. The AI influencer course allows you to generate AI brands that can create content for you, and this opens up a whole world of possibility and even to create AI bands like we saw earlier.
Now, if you'd also like the prompts that I used today, I'll leave a link to that in the description below. But most of all, I would like to invite you to watch another video. This one is particularly interesting because it delves into the topic of censorship in AI video. I think it's incredibly important to understand what are the implications of highly censored video models, and I explore what the situation is, which models are allowing us to have more freedom, and what we can do about it. So feel free to check out that video next. But most of all, thank you for watching. Thank you for being here, and I want to wish you a delightful day.