Top AI Innovations You Can't Miss

As per usual, this has been another week full of different AI releases that you could be putting to work. Today, we're going to be looking at everything ranging from a new tool that analyzes video based on the visuals, all the way to a workflow that upgrades your prompts in a big way. Plus, we'll also be talking about Nvidia's gtc. Welcome to gtc. As you can see, I'm in a travel setup here. This is not my home studio. I'm in San Jose, California right now for the conference, and most of that conference was talking about generative AI. So I have some interesting insights that I want to share with you later on in this video. But let's just kick things off by looking at the brand new things that came out over the course of the last week. Starting off with Describe by Siv Sieve. I'm not exactly sure, but what this does is very interesting. It takes video as an input and then it uses AI to transcribe the content of the video. Summarize that. That'd be pretty usual, but it also uses vision to look at the frames and then take the information from that and then includes it in the summary. So you get an audio visual summary of the video. Obviously it works on the examples here, but let me show you one of mine too. So this is just a random clip from a previous video we created where I show a camera transition where this overshirt disappears. It vanished. Kind of a tricky subject matter. Let's see if it summarizes it. Well, I'm just gonna upload it and say submit. All right, then after 24 seconds, we have a result here. Look at this. In this video, a man with long hair and a beard is captured in a personal space dedicated to music production. Eh, close, huh? Displaying his passion for the craft. There you go. There's further details, but look at this part. A distinctive tattoo on his arm and his casual attire, including a moment where his shirt seems to disappear, adds a relaxed yet focused atmosphere to the setting. So as you can see, it really picked up all the action, including the little detail off the tattoo back here. Now, it does make some assumption as a large language model does with the music production, but I think this is really fantastic. I have to think how this could be usable, but just haven't seen this before and I wanted to share it with you. I think this is a great way of summarizing clips where you're looking for more than just a transcript summarization, but a visual summarization. Very cool app, but now on to the next one. Okay, next up, we're going to briefly dive into the world of 3D. Full disclaimer. This is not my area of expertise. Nevertheless, I want to show you what came out this week. Because this model from Stability AI that produces 3D models from Text prompts is significant. So they released a new iteration of their stable diffusion video model. But now it doesn't just produce video, it produces 3D models. Now there's a bunch of examples on here. There's no space to try this out quickly. And probably the most important note is that if you want to use this for commercial purposes, can you? You will need a Stability AI membership. As per usual. All the links are in the description below. And the biggest difference here to other text to 3D models here is that usually these other models have used image generation to create various images and then they stitch them together. In this case, it actually uses the video model. As you've seen with Sora, that can be very powerful for 3D models because if you have consistency, you can just move the camera around and scan the whole thing. If you want to use this or learn more about it, here's the hugging face space and here's the paper that came out with the release of it. One more super quick 3D innovation. I just wanted to point this out. Stable projectors and this is basically a way to use generative AI to generate textures for your 3D models. This is something I expect that over time it will be implemented into all 3D software. But as of now, you can install this extra application that will help you generate textures for your 3D models. You do need Nvidia GPU for this to work though. And that's really all I got to say about this. Just wanted to point it out. Let's move on to something that more people watching this video will be able to use. All right then. The next one is really fantastic. This has been released by Anthropic and it's a prompt improver. Now look, this is meant for development and yeah, we do need to use a Google Collab workbook to make this happen. But just stick with me if you're non technical, if you don't know code, it's fine. You can use this too. Even if you intend to not use the prompts for development. That's how simple this is. Actually, it's so simple that I'll give you a full tutorial right here, right now. Let's begin. So if you want to improve some of your prompts in a big way, you're going to go to this link, it's in the description below. And then you're going to go ahead and you're going to copy this to your own Google Drive. For this need to be logged in with your Google account, which I am. As you can see up here, save a copy and drive. All right, now I have my copy. I can see this by the title, which says copy of metaprompt. Now, all I need is one thing. I need my Anthropic API key. This does work with the Opus model. And no, you don't need the subscription. You can even be in Europe. This just works. Okay, and the way you're going to get your API key is you go to console.anthropic.com and here I'm going to go to get API keys. I'm going to create a key. I'll call it Demo key one. Copy this over and post this right here in the workbook here, between the quotation marks, I'm going to post the API API key. And also, if this is not clear to you, if you've never used the OpenAI API key, it should be pointed out that the way APIs work is that you're going to be paying per request. And in order to do that, you do need some credits here. So if you go to plans and billing, you can add some money. You should be able to get $5 completely for free just by entering a number. You don't need a US number, although it shows one. So just make sure to have a few dollars of credits in here and then we can proceed. And that is it. We're almost ready to improve our prompts. All you need to do now is go from top to bottom and press the different play buttons. Okay, so I'm going to press this one. We got the green check mark after seconds. Now we're good to go. This took one second. And all I'm going to do is scroll and click on all these little fields like so. Just don't forget to also do this meta prompt one along the way. Press play here, here press play on this one. And this one is going to take longer. Okay. Because the meta prompt is so long, this is going to need some time to go through the whole thing and to improve your prompt. And after exactly 30 seconds, we're pretty much done. All we need to do is click play a few more times. Yeah, it is a few steps, but except that the first one is pretty straightforward. Right? And there you go. That's it. We have the improved prompt prompt Here. Now, this is set up for development, so it does include variables, but this prompt is amazing. Look what it did with this super basic one line prompt. It went from draft an email responding to a customer complaint all the way to this really well crafted opus. Is fantastic at prompt engineering, as I pointed out in the release video. And now what you could do is you could simply replace this part here and run all the fields from here on down. And there it is, simple prompt turned into a seriously effective, precise, unambiguous, high quality prompt. I strongly recommend this workflow. Only downside is each one of these is going to cost you 5 to 15 cents from your API. Matter of fact, I like this so much that we ran a little workshop on this in the AI Advantage community where we help different members in a live event troubleshoot their workflows if anything came up. That's just something the community is really good for and we're going to keep doing that. All right, next up, we have interesting updates from Leonardo AI. As you know, we talked about this before. Fantastic. All in one solution for AI image generation. They have a lot of different models, a lot of different tools, and they added some that are really essential in my opinion. First of all, they added a universal upscaler in beta. We don't need to cover this. We talked about upscalers many times. So that's this result right here. Oh my God. You can basically upscale images with AI where it adds new pixels and uses a generative engine to upgrade the quality of the image. It's not just fancy sharpening, it's generation. And the new feature in here is this little checkbox here. Okay, it's transparency. As somebody who creates videos all the time, this is essential. Generating images with transparency is so amazing. We've seen demos in the previous weeks where you could do it in some hugging face space, but now it's starting to get integrated into apps, which is what makes it usable. Super exciting. So I'm just going to generate this with the free credits that come with every Leonardo account. As you can see, every generation, 20 credits, I get 150 for free. I just logged in with my Google account and basically without touching all of these advanced controls, which you can certainly do, and then you could upscale the image too. It just generates me the image I want. Look at that. A cat with a hat with a transparent background. This is the first on the channel where we got this from app. And what's the point of having transparent images in your videos? Well, I'm just gonna say the magical words of editing Team, show them why it's important to have PNGs of a cat with a hat. Is this a cat in a hat? That's why you need this tool. All right, all right. And to round this week out, I decided that in the end I want a segment where I kind of look into the future a little bit. And this announcement is so big that I really want to talk about it. Apparently, GPT 4.5 leaked. Now, this is not 100%, but under the OpenAI domain. A crawler caught the GPT 4.5 page and there's two big updates which are supposed to come in June, which is kind of far away, right? But one is an increased context window. They're doubling it from 128k to 256 with better retrieval across the whole context. And secondly, the cutoff is supposed to be updated to June. Now, while this does mess up all our hopes for GPT5, it does align with what Sam Altman said in the most recent Lex Friedman interview. So According to that, GPT5 is probably not going to come in the next weeks. So that's our preview into the future. And now I want to talk about Nvidia GTC that happened this week because that really brings all of this together. You might have already seen the summary videos talking about the brand new Blackwell GPU that was announced during the keynote by Jensen. That's incredible. It's going to power the future of this industry and of all software. But I just wanted to share my personal impressions of actually being there and going to the talks, being close to the front during the keynote and spending the week with other AI content creators here on YouTube. And the one big takeaway is it's so damn early. You could see this everywhere. This was a developer conference that was largely focused on enterprise solutions, but it just takes time to build those right. Enterprise solutions don't pop up from one day to another, and everything we're looking at here are either demos or their consumer apps. But it's so early with all of this. Could really see this at the conference, all this new technology coming out, the hardware, the software that is being developed from Nvidia side or other vendors. It just takes one to two years for some of these solutions to be implemented. And it just hasn't been that long since the ChatGPT moments. So you just have to respect the fact that what we're looking at here is early adopter type of stuff. And the big apps, the implementations like Copilot, inside of Excel, I mean, they're just releasing these things, implementing them for the first time, feeling them out. So if we're talking use cases, there's just going to be more and more over time because it's literally present in every industry and we're just getting started. So, yeah, the conference was a blast. Great job, Nvidia, for bringing together all of these amazing people into one place for multiple days. Fantastic experience, and I've never been more optimistic about the future of this technology. This is moving at a pace that nobody would have predicted. And I'm glad to have you on board for this wild journey where we explore use cases week after week. That's everything I got for this week. I hope this was helpful to you. We do this every single Friday, so make sure to subscribe to a new episode like this every single Friday. And if you want to discover more use cases right now, check out the full playlist with past episodes that include everything that came out over the course of the last few weeks. All right. For me, it's time to pack up this mobile setup. I'll see you around.