Insane AI Advancements Keep On Coming!

With the insanity of all the announcements over the last few weeks in AI. So far this week has been relatively slow in terms of announcements, but that doesn't mean there isn't some really, really crazy stuff happening behind the scenes in AI. In this video, I want to walk you through some of the craziest stuff that we don't quite have access to yet, but is just right around the corner. So stick around, because this is some really, really cool stuff that's in the works right now with AI. So the first thing I want to show you is this research paper that recently came out called make it 3D high fidelity 3D creation from a Single Image with Diffusion prior. Now, I know that's a mouthful, but basically what that means is that you can take a 2D image and this will convert that 2D image into a 3D object. There's been a few Google collabs and a few hugging face spaces where you can upload an image of a person and it tries to make that person 3D. There hasn't been much in the way of making objects 3D yet. Here's a picture of an astronaut, and you can see the astronaut was converted into a 3D object. You should be able to take that object into a tool like Unreal. Here's a picture of a blue jay sitting on a basket of cookies. You can see how it converted it into 3D. Here's a picture of a bunny sitting on a stack of pancakes converted into 3D. And all of these objects you can import into places like Blender or Unreal Engine and use them as objects in in games or 3D videos or whatever you want to do with them. Here's some more examples. A Picture of a 2D image of a house turned into a 3D image. A 2D image of a gorilla turned into a 3D image. Here's a picture of a frog holding a birthday cake. A picture of a teddy bear, a skull, a castle. You can see how it took just a 2D image and then figured out what it would look like from other angles and then made a 3D object out of that image. You here's a DeLorean and a motorcycle, a cactus, some sort of bunny statue. I believe that's the Sydney Opera House. A corgi wearing a top hat, a squirrel with a ball, a knight on a horse and a hamburger. You can see, basically what it's doing is it's taking a reference image and then generating what's called a neural radiance field from that reference image and a Depth map from that reference image and then sort of guesstimating what it would look like from various angles. Now, if you really want to dive into the math and how this actually works, check out make it 3D GitHub IO and there's some great explanations and an explainer video here that tells you how this technology works. Now, the next one I want to show you also isn't available yet, but you can see what it looks like over@editavideo GitHub IO. This one's called Edit a Video Single Video Editing with object aware consistency. And you can see it uses a reference video and then a text prompt to change the reference video to the text prompt. So, for example, a bear playing the guitar. From a person playing a guitar, change it to a monkey playing a guitar. Change it to a man playing a guitar in a cartoon style or a man playing a guitar, Matisse style. We've got a reference video of a woman on a swing. Change it to a man on a swing. Change it to Iron man on a swing. Change it to a woman on the swing, Van Gogh style. Or a watercolor painting of the woman on the swing. Here's a picture of someone playing basketball. They change it to Lionel Messi dribbling a basketball, or a raccoon dribbling a basketball, or an oil painting of a man dribbling a basketball or a man dribbling a basketball, Renoir style. If you've watched my channel for a while, you know I suck at pronunciations. Here's one of somebody running as the reference video. And they change it to a zombie running, a werewolf running. A man running, Pixar style. Pencil sketch of the man running. Here's a cow changed to a bull walking, or a zebra walking, or a cow walking in the snow, or a cow walking in the desert. Now, we have seen other stuff like this where there is video reference and then a text prompt to change the video. But here's a comparison of some of the other ones we've seen so far. So here's the source video of a man boxing. And then here's the text prompt. A Batman is boxing. So this is what it came up with using this edit a video style. This actually looks like Batman boxing. Some of these other versions that have come out, you could tell, are a little bit funky. This one, I don't know what's going on. He's got like a little ear coming out of his head, but it doesn't really like Batman. This one kind of changes from frame to frame and is A little funky. And this one has barely any reference to Batman at all. Here's another example. A gorilla is riding a motorcycle. Here's the source video of just a man on a motorcycle. And here's the edited video version of a gorilla on the motorcycle. Here's the frame wise version. Not much of a gorilla there, Tuna video. You got kind of a cartoon gorilla, but it really changed the entire video. It changed the background, the color of the motorcycle, everything. This version made the motorcycle, the background look consistent. The they just put a gorilla on the motorcycle again. If you want to learn more about this one, check out editavideo GitHub IO and you can find the entire research paper here and really dig deep into exactly how this one works. Now the next one I want to show you I came across from the rundown and check this out. This is an iPhone tool where you can point your iPhone at anything in the real world and it will actually give you the hex color of that thing in the real world. So let's go ahead and play this video and take a peek. He points it at his computer and you can see the hex code change right here on his iPhone based on what he's pointing it at. So you can literally get the color of anything you want as a hex code that you can use on your website or your Photoshop or anywhere you need to figure out the color for it. See, took the plant, figured out the hex code for the yellow plant there. This is called Color GPT. They do have a GitHub page. If you search out Color GPT you should be able to find it. But here's the URL. Now this one you actually can use if you want. If you go to their GitHub on your phone and when you're on the GitHub page, come over to the right sidebar and click on colorgpt Vercel app. Now if you do it from your iPhone here, you should be able to press this button that says capture color, allow it to use your camera, and then it asks you to enter the API key for OpenAI. You can create a new OpenAI key here, copy it, jump back over to the website, paste it in here, and then save the API key. Now it is sort of a pain in the butt. Hopefully they have an actual app for it soon. But as you can see, I could now select any color in the real world. Look, there's my recording software, there's my salmon colored shirt, there's my orange bar at the top of my mixing tool. There's my Coke Zero, can my keyboard, my camera color. And you can capture any color you want and get the exact hex code. And this works right now straight from within any phone, as long as the phone is connected to a browser, has a camera, and you plug in your API key. So again, that's called Color GPT and you can actually go directly to it on your mobile phone through colorgpt Vercel app. So the next tool I want to show off, we do not have access to yet, but it's called Dreamlab gg and this is to create animated game assets. So check this out. They enter a prompt here, person wearing white biker jacket, walking. A girl with red hair, black jacket, facing the side, walking. And is creating this whole animation here. White shirt, blue jeans, footwear, sneakers, walking, boxing, fighting stance, facing away so you can see. It will create these animated game assets that you'll be able to use inside of a game or a video however you want. And it is wait list only. I don't have access to this one yet, but it looks like it's also got features where you can build game levels, where you build this level here and then you can just drag and drop the assets onto the level. And these assets here are created with Stable Diffusion. See, they just click generate new image robot made out of stone, friendly guide NPC 2D game asset. And then they click generate. And then it generated with Stable Diffusion this robot, it added the robot here and now they can drag the robot straight into their game. So it's just a really cool drag and drop game editor where you're creating the assets for the game directly inside of Stable Diffusion. You can see right here it says which image generation methods are supported. Stable diffusion 1.5 and 2.1. You can bring your own models, loras and embeddings and more. So it looks like you can even have your own custom train Stable Diffusion models to create game assets with. Now, according to their website, it says, what's the release schedule for Dreamlab? And they say we're rolling out our animation tool this March to everyone. Level editing and multiplayer capabilities will be available in April. We're getting pretty close to the end of March right now, so I would imagine it'll be coming out pretty soon. They do have a discord where you can stay up to date on it. This is something that if they can keep the deadlines for and actually roll it out soon, this will be really, really awesome. Because one of the things I want to experiment with in future videos is trying to Create my own game. And this is going to shortcut the process for creating animated game assets. So this one's called Dream Lab. You can find that over at Dreamlab GG and check out their wait list. I'm really excited to get my hands on that one. All right, so the next two things that I'm going to show you, I think are really going to blow some minds. I don't know how many people watched the south park episode with ChatGPT where the kids were all texting with their girlfriends using ChatGPT to make them sound more caring for their girlfriend. Well, this is like that on steroids. This person combined GPT4 whisper and AR glasses to create something to generate better conversation. This tweet from Ryan HP Chang here says, say goodbye to awkward dates and job interviews. We made rizzgpt Real Time Charisma as a service. It listens to your conversation and tells you exactly what to say next. Built using GPT4 Whisper and Monocle AR glasses. This is pretty crazy. Here, check this out. Hey, Varun, I hear you're looking for a job to teach React Native.

Thank you for your interest. I've been studying Reaction Native for the past few months and I'm confident that I have the skills and knowledge access necessary.

I mean, as he's reading it, it kind of sounds like he's reading off of a teleprompter, so you probably want to get a little bit smoother with that. But as you can see inside of his augmented reality glasses, it's listening to the conversation and then telling him how to respond to the conversation in real time. This is sort of a fake simulated job interview here. Gary.

For the job.

What do you know about the Use State Hook in React Native?

I know that the U State Hook is an innovative and rapidly growing software engineering hub that specializes in React Native.

I mean, if I was conducting a job interview and somebody was responding to me in that way, I know that Joe React Native is this. I would probably see some red flags. But we're getting to a point now where you can have a teleprompter in your glasses that are listening to your conversation and then telling you what to say.

Next development I'm confident that I can contribute to.

Sounds like you're really great for the job.

What else do you want to tell me? Thank you for your confidence.

You can also tell they sped up the video a little bit because the GPT4 behind the scenes probably takes a second to process what it just heard. So there'd be long awkward Pauses at the moment, but this shows you where the technology is going.

I've done extensive research and practice with the U State hook and I understand its core principles and concepts. I believe I can use this on the job.

Thank you. Sounds like you're hired. So that. That to me is pretty mind blowing. Not something that we all have access to yet. They definitely hacked a few things together. Here's what he actually looked like in the conversation during this thing. You can see he's got this big old thick thing on his glasses here. Here's a close up view of the augmented reality thing that was hanging over his glasses. I mean, that might be a little bit of a red flag if you had that hanging over your glasses while you were actually doing an interview. But it's probably not very long before we're just wearing glasses that look like these glasses here. And this whole gadgetry is just built into the glasses. I mean, that is really, really close. I'll link to this Twitter thread down below. Because if you scroll through the Twitter thread, they actually do have a little bit of an explanation of how they built it and how the whole thing works. So if you want to do a deeper dive of exactly how this was created, they do show you this on their Twitter. All right, so the last I want to show you is called genmo. Now genmo is a video tool and I did cover it in a previous video. And in that previous video it basically was a text to image thing where you had it generate like a fairly static image and then it would do almost like a deforum looking video. Well, this new one is pretty groundbreaking because you can do all sorts of stuff. You can create new images with it, edit existing images, animate images and and actually generate text to video directly inside of genmo. So this one's available at genmo AI. This version that I'm showing you right now is wait list only. It's not available to the general public, but you can get on the waitlist and they're gonna be rolling it out slowly and giving more and more people access. This is an early access version so that I can demo it for you. So this is using GPT4 behind the scenes and I can have it create an image for me. So create a photo realistic image of a city skyline at sunset on a cloudy day, Highly detailed, dramatic lighting. And let's see what it generates for me. All right, so it generated this futuristic looking skyline for me here. Now check this out. Animate the clouds in this image. Sure thing. I'll animate the clouds in the image. How many frames would you like the animation to have? Let's go 15. Check that out. It animated the clouds. You know, you can tell it's kind of messing with some of the skyline as well, but it's actually not messing with anything in the lower portion of the video. All right, so check this out. I can actually upload it my own image here and animate my own image as well. So let's go ahead and click this image button down in the bottom corner. I've got a cool image of an astronaut here. Let's go ahead and open up this image. I made this one in mid journey version 5. And you can say, great. What would you like me to do with this image? Let's just say make the Earth in the background rotating. And then it says, I can animate the image to make the Earth rotate. How many frames would you like and how much dynamism and exploration? I don't really know. Let's just say you decide. All right, I'll animate the image with 45 frames. 35 dynamism and 45 exploration. Personally, I don't totally know what those terms mean, so let's see what it comes up with for us. It's kind of funny. It made an animation where it turned the astronaut's face into the Earth. Not quite what I was anticipating, but, you know, definitely an interesting animation. You never know what you're going to get with some of these AI tools. Now here's one that I did earlier when I first got access and started playing with it. I uploaded this image that I made in midjourney here. And what's really cool about this is it can actually read what's in the image. I uploaded this image and it says, an image of a house sitting on top of a hill next to a forest contains objects including cloud, sky, atmosphere, plant, building, dust, sunlight, afterglow, sunrise, atmospheric phenomenon. And then when I upload the image, it says, great, what would you like me to do with this image? And I said, animate the clouds in the sky. And here's what it generated for me. You can see it left the house and the hillside untouched, but now the clouds are just animating throughout the sky. But while everything else stays static. Now, genmo also claims that it can do full on text to video. So if I click New Chat here, there's an option to generate video. Genmo can create videos from text or combine clips into a movie. So if I was just to do something like create a video of a monkey on roller Skates. Let's see what it does. It asks me, would you like me to generate a square video with the caption a monkey on roller skates cruising down a sunny boardwalk with palm trees swaying in the breeze? Sure. And there we go. We got a monkey skating on the boardwalk there. Now here's the demo video that genmo shared when they announced that they were launching this new chat version. You can see they typed, help me direct a short movie, Snack to the Future. Genmo says, sure. How about a story where Doc Brown and Marty McFly take a side job delivering for Door Dash. They use their time traveling DeLorean to live for food in record time. But they mix up an order with the help of a quirky inventor. The driver must ensure historical figures receive their correct orders without altering the timeline. They respond, great. Make some video clips. And then genmo responds, I'll generate video clips for your Snack to the Future movie. Here are the captions I'll use for each clip. A cinematic shot from Back to the future of Doc excitedly showing off his time traveling car to Marty. A vibrant red food delivery bag traveling through a wormhole. Doc and Marty delivering food to Abe Lincoln narrowly escaping a t. Rex. The DeLorean engine exploding. DeLorean flux capacitor with sparks flying. Doc and Marty jumping for joy after fixing the DeLorean. Do you want to make any changes to these captions? Let's end with a sunset. Great. I'll replace the last caption with the DeLorean driving into the sunset. And then it goes off and starts generating all eight of these clips. Once it's all done, here's the movie it generated. Snack to the Future. There's the DeLorean. Doc and Marty join forces, deliver food to the past. There's their food delivery bag. There's their wormhole. Doc and Marty create a paradox by delivering tacos to Abe Lincoln. There's them wrestling a dinosaur. Uh oh. There's the DeLorean smoking. They fixed the flux capacitor. There's the flux capacitor. And there they are driving into the sunset. The end. Now, obviously this is super early. This is the worst it's ever going to be. It only gets better from here. And this tool is pretty dang crazy where you can just have a conversation with the chatbot and through this conversation generate videos. You can learn more about this one over at genmo AI. And if you like learning about all of this future tech and all of these cool AI tools and you want to stay in the loop with everything that's coming out and everything that's about to come out. Head over to Future Tools IO. This is where I curate all the coolest tools that I come across. And if it's a little too much, you just want the TLDR of the Week. Join the free Newsletter Every single Friday, I'll send you an email with just the news that you should know for the week, as well as just the five coolest tools that I came across, a handful of YouTube videos, and one cool way to make money with AI. It goes out every Friday. It'll keep you in the loop with all of the crazy advancements that are happening in AI right now. Totally free to join. You can get on the list at FutureTools IO. Thank you so much for tuning into this video. I hope you enjoyed all of this stuff. Lots and lots of cool stuff coming. So far this week has been a little bit slower of a Newsweek in AI, but. But that doesn't mean that the advancements and all of the crazy stuff going on behind the scenes in AI isn't still moving at a rapid pace. As you can see, there's lots of crazy stuff happening right now in AI that we're just on the verge of getting access to, and everything we saw today is still super, super early. Some of these things are just the research papers and just the original experimentation of it all. What we're seeing right now is the worst it's gonna get. It's only gonna get better from here. All this stuff only gets better and better and better. And as people get access to these tools, iterate off of them, build on top of them, and make them better and better and better. So really exciting times in AI. If you like this stuff and you want to stay on top of it, click the little like button. That'll make sure you see more videos from me in your feed. And also, if you haven't already, click the subscribe button because that helps me have more subscribers on this channel. All right, thanks so much for tuning in. I really appreciate you. See you guys in the next one. Bye, Sam.