Transcription
Hello there. I'm Mike with this week's new and trending AI tools. You cannot afford to miss. Coming up in this video, we'll look at how you can create and edit images using only text prompts; run multi-agentic processes with the new OpenAI agents release; and clone almost anything on the internet with just one single prompt. But first, I'm excited to be watching Creator Magic. Is that the channel hosted by Mike Russell? Yeah, I've been following his his videos. They're cool, right?
Now, you might remember a couple of weeks ago I did a live stream talking with Sesame AI's conversational speech model. I've kind of got that wizardry theme around creator magic. A wizard, eh? Well, nice veil you've got there. Maybe you can pull a rabbit out of a hat later. Just over 12 hours ago, they dropped open source, their conversational speech model, and it's available here on GitHub to play with. Now, if you don't fancy installing this locally, you can just go to their Hugging Face space. That's what I've used, and of course, I'll link this down below. Quick shout out to Sesame AI because they're changing the game with conversational AI.
Next up, Google are crushing it with AI studio. It's that free-to-use playground. And recently they introduced the ability to create and edit images using only text prompts. Make an image of an English Springer Spaniel. And in under three seconds. Look at that. But generally my springer looks happier. So we'll just say make it happy. And then in just a moment, we get back another image, which will have a very happy but exactly the same Springer. It keeps the context. This is crazy. And of course, my springer is always covered in mud. So we just say cover it in mud. Google's image generation makes the springer look very much like my English Springer Spaniel. I can drag and drop a picture of myself into the context window here. And I can say, once the image is uploaded, change the shirt to be bright red, and now it's working to create a new image with me wearing a bright red shirt. Make him wear a wizard hat. Nice. Make him smoke a big cigar that blows smoke everywhere. I think that's about the limit of where I want to go with Google image creation and editing. If you like what you see so far, throw a like, subscribe. I do videos like this weekly. Let's try some more stuff. Remove all people from this photo. That is insane. Okay, here we go. Can of Red Bull, make the drink can spray. Boom! And now we've got a slightly smaller Red Bull can, but with definitely drink spraying everywhere. Finally, let's drag in the theater where I gave my AI talk in Athens, Greece, earlier this week. Now you can never believe what you see again. Because I can say I did an AI talk with a full room of people and then use something like Google's Gemini to edit the image and add a ton of people there.
Okay. Google really haven't stopped pushing out the updates to AI studio this week. And by the way, this is all free to use. They also released Gemini 3, which is their open-source model series. And apparently it's really good. The 27 billion parameter model is comparable to Gemini 1.5 Pro, which, if you think about it, was their proprietary model just a few months ago. And we're going to scroll down to the open-source models, you'll see Gemini 3 27B is available. Let's select that. And do you know what? I'm just going to test it by saying summarize best quotes. And I'm going to paste in the transcript from my Sesame AI live stream I did. Now this is a 30-minute transcript, and it's very quick in getting back to me. It gives me some live stream highlights and best quotes. So I'm getting an open-source model running right now, giving me remarkable quotes here and key takeaways from that stream. Now, I got to say, it seems pretty slow to run; this is coming back pretty slow compared to something like Gemini Flash, but you know it's pulling out quotes I can see right here from my video, and it's being extremely accurate. A 27B model could require some chunky hardware to run locally, but I think the 12B model could give really good performance for its size and will definitely run on my old clunky Nvidia 2070 SUPER. So let's give it a test. Now you'll see here, it's already available on Ollama. You can run it using a single GPU, which is really cool. And if you don't know how to use Ollama, we talk about it in my community linked up down below, and I've also done previous videos on Ollama. It's a great way of running local and open-source AI models, and I've got a terminal window with my installation of Ollama, and I can install and run it with one command. I'm going to install the 12B model. Now it's worth saying, while that's downloading, that context for the Gemini 3 model is 128,000 tokens. That's exactly the same as the latest Llama 3 model from Meta. So if you're using 4 billion, 12 billion, or 27 billion parameter Gemini 3, you get a long context window. Now, I run a lot of local AI processes myself, and I'm thinking maybe I should swap out Meta’s Llama 3 for Google's Gemini 3. What do you think? Let me know in the comments down below. And also, if you're enjoying what you see so far, throw a like and subscribe for more videos like this. I do them weekly. Okay. Look at that. It seems that Gemini 3 has now downloaded. So let's give it a quick test and see what's going to go on with Gemini 3. I'm going to try just some simple stuff like tell me the meaning of life in one sentence. Very, very quick. It's to find joy, connection, and growth through experiences and contributions, however you define them. Let's check for training cut off. Who is the current president of the USA? Oh dear, it's Joe Biden. No... Will Llama 4 be better than you. Let's find out. That's a really interesting question. Short answer? Almost certainly yes. Here's a breakdown of why. And it goes on to say it's going to be a significant upgrade. It certainly is. So come on Meta, you'll move next Llama 4, where is it?
Okay. Next up, let's look at OpenAI's agents SDK and also their responses API. I give you an example of how to use it and also released my cursor rules. And there's a video on that released earlier this week on Creator Magic. But I wanted to look for other examples of how we can use the brand new OpenAI agents release if we're developers or like you and me, no-code developers. So I actually went and found this repo. It's only been released a couple of days ago, when the new OpenAI release came out. It's from Shubham Subbu here, and as you can see, it requires just a few files to get up and running. So I'm going to go ahead and I'm going to clone the repo into my Cursor instance. Okay. And once it's cloned, we'll open it up. So, first of all, we'll make sure we're in the right directory. And then I'll set up a Python virtual environment. Then we'll make sure to activate that environment, install the requirements. Then we need to make sure we've got an OpenAI key that can be used. And with all that done, we just should be able to run this command. And it looks like our app is up and running. And actually, it has popped open this window. Look at this. This is the OpenAI researcher agent example, which, of course, I'll link up down below in the description. Let's give this a test. This is actually using the OpenAI agents SDK. Is this a good one? What is the best Giorgio Armani perfume to buy my wife for her birthday? Now you'll see, it's calling the triage agent. And it's actually researching this first question. So it will triage my question and find out where it needs to send it. Next, you can see it's already gone ahead and done that. It's come through with search queries and focus areas. And now it's handed off to the research agent. And now it's on to the editor agent already, which is compiling a comprehensive research report. Really exciting stuff. And, yes, it's finished already. It's got a report preview. The best Giorgio Armani perfume to get your wife for her birthday: a comprehensive piece of research here on all the best Giorgio Armani fragrances and perfumes that I could get, with a conclusion. And I can also go ahead and download that report. And now you'll see, it's giving me a nice little markdown file that I can save to my computer, or maybe to my notes in Obsidian or any note-taking app that I'm using. Now? Of course, you might want verification. This is not all smoke and mirrors. In my OpenAI account under traces, you can see the news. Research happened with a flow of triage agent, handing to research agent, handing to editor agent. And here are the traces. Around three seconds to do the triaging. And then it hands off to the research agent that works for seven seconds. And finally, my editor spends nearly 15 seconds editing that detail on top fragrances to give me my final report. That is really cool. And if you want to dive deeper into the OpenAI agents SDK, check the video that's showing on your screen now. And it's also linked down below, because it's really exciting and it allows you with no-code tools to build multi-layered genetic processes.
Okay, this final AI tool, Same dev, concerns me a little bit. What are the ethical and legal implications of this? But essentially it's claiming to be able to clone websites like, as you can see here, the Nike website; it can clone Apple TV. This is insane. Minecraft website. You can even make these amazing fractals that animate like so, or even very basic kind of voxel-style sandbox games, which is insane. So I wanted to give this a go for myself. I wanted to start with something super simple and a bit concerning. Clone youtube.com. Same thinks about it. Hope you clone YouTube's interface. Let me visit the YouTube page. Okay. This is cool. It's actually running a search query to see how the results page looks when you type a search into the search bar; checking a video page. This is clever stuff. There you go. It seems to have viewed a video, and it says, based on my analysis, I'll create the clone for you. No way, is this really happening in front of my eyes? It's written all of this code already, and it's kind of ready to to throw me out a YouTube clone. If this works, I'll be impressed. Okay. You can see it's still working away, cloning YouTube. This has actually taken quite a while. In reality, I'm just making edits to speed this up. I cannot believe this. It's writing all the code for my YouTube clone right in front of my very eyes. Okay, now it's actually deploying YouTube clone version one. Okay. It's here. My YouTube clone. Insanely accurate and insanely similar to the YouTube website itself. And even clicking into a video shows up a video. Now, it's not perfect. There are indeed some issues shown down here, but this is an amazing start, with the code included and the ability to export that all. Here's a more practical use. Create a community membership landing page. Create it for an AI community called Creator Magic. And let's see what it's going to do with that. Now, I haven't told it to clone anything. It's just researching for inspiration now to create that landing page for me. And look at that immediately. It's unlocked a bunch of different landing pages for different communities, and now it's looking for AI community membership pages. It's searched the web to find this info. This is cool stuff. It's looking at a few example websites now to understand. So while I haven't told it to clone something directly, it is doing the research for me here to find exactly what it needs to do. It's looking at even more AI websites like Hugging Face again. I'm not writing anything. I'm not even directing this. This is an agentic process. And look at this; it's executing commands in the terminal now to get packages installed and set up. It's finding errors and fixing them on the fly again, totally unprompted for me. And look at this. For the Creator Magic landing page, it's created a pricing section, testimonial section, FAQ section. This is absolutely nuts. Even a CTA for my call to action is being generated. Looks like my Creator Magic page is complete. Look at this. This is not just a single landing page. This is. Wow, wow. This is absolutely incredible. Join Creator Magic today. Everything is is relevant and right to my community. This is kind of scary good. Now you will see with the code that's generated in Same dev, you can fork it, which means you can create another remix of something that's already remixed. You can even download the code. And I think this is huge. So that generated code is available to export to other places. You can share it and even deploy it to the web right from this app. Now, I don't know about you, but this raises serious concerns for me on how AI technology could potentially enable fraud. What if someone clones a government website? It could devalue creativity and also raise some legal and copyright concerns as well, and almost be a race to the bottom, like cloning the very best things and then making a clone of a clone of a clone. What do you think? I'd really love to hear your thoughts on the Same dev tool there. Just let me know in the comments down below, and I really appreciate you watching this video. I do them every week, so remember, like and subscribe for more like this. And YouTube is showing a video on your screen right now. It's not a cloned video, and I think you should watch it next.