NEW AI MODELS! Nvidia eDiff Demo / DALL-E 3 / Anime Models – AI NEWS

Oh, Jesus, who hit the record button? I wasn’t ready. What’s this video going to be about? AI news? AI news? Hello viewers, welcome back to another video. Today we’re going to be discussing the latest and greatest in AI news. Now again, this is a nice curated video for you guys. But if you want to see AI news in real time, I really strongly suggest you go down in that description, join my Discord server. I have a whole channel dedicated to posting the latest AI news. And you viewers, you subscribers, you Discord members do a fantastic job at just dumping the latest and greatest in that channel. In fact, that’s where I get most of my AI news for these videos from that channel. Let’s jump right into the AI news, why don’t we? In video released, their own text to image model. All right, you guys were pinging me about this. You guys were telling me, Matt, you need to make a video on this. Yeah, I know guys. I know I saw it too. I wanted to wait to put this in AI news though, because there’s no demo. Yeah, that’s right. They released an AI. They just showed it off basically just dangling over our heads saying, hey, this is really cool. I bet you wish you could have this. Oh, you can’t have it. But the paper is honestly really intriguing. They show us a few different prompts here. Basically it’s like a few different diffusion models mush together in one and it does like upscaling. It seems to operate very similar to something like Google’s party model, which means yeah, it’s going to be good. And it definitely seems to be good like here, a highly detailed digital painting of a portal in a mystic forest with many beautiful trees, a person standing in front of the portal. Pretty good. Pretty good generation there. I feel like we could get that out of mid-journey or dolly though. Highly detailed zoomed in digital painting. Can’t dress as a wish wearing wizard hand haunted house art station. Yeah, pretty good detailed image here. And something I could see coming out of dolly image of a beautiful landscape ocean. Again, this is just a nice pretty picture, something I wouldn’t be surprised to see out of other models. Although this also offers a few different methods for generating that we haven’t seen by other models before, such as the ability to paint and put words into different objects and then have it create an image. And they call that paint with words. So we’re going to take a look at that demo in a little bit. And I’ve also got like style transfer too, which I’ve seen before on other websites we talked about it with night cafe. Style transfer is a real thing. And they show style transfer right here. This is the reference photo and then they transferred it onto this prompt of like a duck with a skateboard. And it worked out pretty well. A lot of people were like, oh man, is this going to be the future of AI? And I mean, yeah, I would agree. I also see Google Imaging and Party doing very similar stuff. And I have no doubt that the other big companies are also working on stuff like this. You’re going to notice that mid-journey is not tested in this paper. It can synthesize samples of 64 by 64 to 56 to 56, 1024 to 1024. The one thing they don’t really talk about unless I’m missing something is the performance of this model on a large scale that’s really important. You can’t really make good money as a business on these models unless they’re efficient. That’s why stable the fusion is so good. If you try to run Dolly too at home on your computer, I like 99.9% of people would not be able to run Dolly too on their computer. It’s just way too much requirements. But stable the fusion on their other hand can run on a lot more of people’s computers. And mid-journey clearly is pretty efficient too based on their pricing where you can get unlimited generations for 30 bucks a month. So yeah, like sure this model could be really good. But on the other hand, if it takes a lot of processing power, what’s the point at the end of the day because it would be way too expensive for everyone to use? Anyways, though, we’ll stop being doom and gloom and take a look at what this model is capable of. Obviously, it’s denoising. These are diffusion models, starts out with random noise and turns it into an image over time. Here are the comparisons that they provide for us. Stable the fusion in Dolly too, where the ones they compared it to, which is pretty interesting. Stable the fusion definitely deserves to be on there as well as Dolly too. But we don’t see any mid-journey V4 on here and I don’t know if this came out before mid-journey 4 was released. It was like right at the same time, but they just did not compare it to mid-journey. I would be intrigued to see this compared to mid-journey V4 though. Could be a video for another day. Anyways, two Chinese teapots on a table one pot has a painting of a dragon. The other pot is a panda. And obviously they show that theirs can do the teapot with a panda and the dragon where these other ones are just two teapots on a table pretty incoherent. We don’t see the dragon or the panda on them painted well. This could just be cherry picking though, like how many generations did it take for them to get this really good one, for example. And maybe Dolly too could pick that up? I highly doubt it though. This is a pretty complex prompt. I don’t think Stable the fusion would ever get it. And again, you know, the models that aren’t in video’s edif get confused where they just make like bears hanging out with bears and then this one knew it was teddy bears supposed to be on that specific bear, for example. So just demonstrating how knowledgeable it is. And in this one, obviously it’s showing that it has the ability to spell words out onto a t-shirt, which is something that Google’s party can also do. And they show that again where, you know, Stable the fusion is never able to do it. Dolly too can sort of make words out, but Nvidia’s edif can actually spell. And you know, while it can spell, I gotta say like in the background here and these monkeys are a little fuzzy. They’re a little fuzzy. I’m not gonna lie. I’ve looked that much better than the monkeys that we see with Stable the fusion. Oh yeah, Dolly too. I don’t know why there’s dogs instead of monkeys. That’s just weird. Very, very strange. They got the spaghetti with the sushi on it. Obviously they’re showing that Nvidia’s edif was the only one that was able to get sushi on top of the spaghetti. You know what? Let’s just run this in mid-journey like really quick. I’m just curious. All right. So remember, edif was the only one that could get sushi. There is mid-journey’s results, very, very similar looking images, super close there, not really seeing that much sunch. But we got a little sunchies. Little sunchie maybe right up there, getting close, getting close. Mid-journey’s very good is my point now. So I think mid-journey with updates in the future could be as good or better as Nvidia’s edif. And then they have this one of a close-up deal, so our photo can’t writing a scooter. Actually, all pretty decent for these bandana around its neck. You know, Dolly too was able to do it. They’re saying, you know, edif was the best out of the bunch. I think that Google’s party or image in would probably be better than all of these to be honest in terms of like the coherency aspect, which is really what Nvidia is trying to show off here with edif. But nevertheless, very, very good model, I think. And honestly, the shining part of this really is those style transfer and paint with words that Nvidia was talking about. This is definitely some of the best style transfer I’ve ever seen. And they show off some paint with words demo where you just pick out some brushes with different prompts and each brush is like a prompt and it can just make a really nice image. I think that’s a really cool and interesting way of painting, using text to image generators. And I’ve never seen a painter this advanced. I’ve seen some other websites that do somewhat similar thing like a collage method, not necessarily just a painting method. But this really takes the cake where each individual brush is like a prompt or a word. And it does honestly do a really, really good job with this. There really is no word on when we will be able to access this thing. But considering they made an entire demo, I’m really, really hoping that they’re going to let us play around and mess around with us. And Nvidia is a pretty big company too. So I think they’ll let us play around with us. So next up, this is a smaller little bit of news, quarter digital, which is a pretty big channel in relation to VFX. They’ve been messing around with AI a lot lately. And they made this video called Spider-Man, everyone’s home. And they basically use some different AIs. A lot of AI tools actually and some VFX to combine it into this entire video where like all of the Spider-Man’s ever meet up. And you know, they, you could see they’re doing like this cartoon style on all of the different Marvel superheroes. And it’s only a three minute video. So I won’t really spoil much of it. But I thought I would mention it and link it in the description. It’s a really cool use and demonstration of AI tools and what you’re able to accomplish with them, like making Thanos, the Marvel character into more of a cartony. Look with the direct footage from the movie, as you can see. And yeah, they just do a really, really good job of this in general. It came out really, really cool. It’s a great demonstration of the power of AI technology, I think. And how fast and far it’s come in just literally under a year. So next up, I’m finally officially talking about Nege journey. This is a anime based model that is directly working with mid-journey to create anime art. Eventually, it’s going to be offered in the mid-journey discord, I think, or in a separate Nege journey discord. Either way, it’s, it’s really very similar to mid-journey, but it’s all just anime based, anime based, anime based, I teased it and talked about it in a few other videos. But it’s, it’s really, really good. I’ve been testing and using it. You can sign up for the wait list with the link in the description. As you can see, it’s mid-journey X Spell brush. So those are the two companies working together. And yeah, you got to join the wait list here. Thankfully though, I already have an invite so I can do a little testing for you guys today. Just poking through the discord. You can see some of these results are just absolutely fantastic. This was just 8K masterpiece cathedral and it’s really capturing an anime style when creating this image. It definitely looks like a cathedral looks drawn. It’s really, really good. It’s got the nice color palette and the right amount of bloom on certain areas. It looks like an anime image of a cathedral for sure. Obviously, the model’s ability to do characters is going to be really important here. So this is Mr. Bean, which is pretty funny as like an anime character. And it did a pretty good job capturing Mr. Bean and what he looks like in his features and applying it to an anime style. Specifically, this was Mr. Bean on a bean bag eating hot beans and saying cool beans. So I think it did a pretty good job given the prompt. And as you can see, just character design in general really seemed to be Nishi Jurni’s absolute cream to le cream, if you will. This was Fant Griffin in bands of the Hawk armor from Berserk. I don’t know what that is, but it definitely did a Fant Griffin in armor and it did a really, really good coherent job at capturing what this character might look like, what his outfits are, the different colors of the character. It did a fantastic job in general for sure. And yeah, obviously it’s going to be really, really good at your more basic anime stuff, which is probably what you viewers at home would really be using it for is create like these little anime characters and stuff. Obviously, it is just absolutely ridiculously good at this. I mean, the character designs get crazy good like straight up at the level of a human for sure. It doesn’t get much better than this in terms of AI generated anime art. A sign up for Nishi Jurni in the description. It’s hopefully you guys will get in soon to the discord. It’s really, really cool for sure. Moving into some other news here, I’m mad who is the CEO of stability AI, the company that made open source stable the fusion, the text image generator, they want to do a papers with code, the Lactica demo, of course, Galactica is the science text AI that I covered actually my last video by meta AI. They want to put this thing back online because meta AI ended their demo of Galactica. And E-Math thinks that it is so important and impressive for AI as a whole that the demo needs to go back online. And I agree honestly, it wasn’t up for that long. We didn’t really get to experience and mess around with Galactica at the full scale that we should have. So the poll there was overwhelmingly yes. As you can see, I also voted yes over 3000 votes put Galactica back online. I thought this was a cool little piece to include. And yeah, go check out that video about Galactica. It is really, really a fascinating text model all based around science in other E-Math related news. Imagine something very interesting in this interview here, which I’ll link down below. E-Math, stable the fusion stability AI and what’s next. A viewer tips this off to me. So we’ll just take a listen at what he has to say. Nobody behind, but the reality is that open source will always lag close source. So like mid journey just released version four, which is amazing, right? And dali three will come out soon, which will be even more amazing. Why because they can take open source basis and so he’s basically talking about this in the context of how open source will always be a little bit behind closed source projects which makes sense. But he says that mid journey V4 is really, really good. Of course, we all know this by now. I made a whole video on it. But he says, Dali three is coming out soon. He just says that it’s going to be out soon, which is really interesting because I just made a whole video where I talk about why I think that’s not going to happen. But obviously this guy is the CEO of stability AI. He knows a little bit more about these things than I might. So he says, Dali three is coming out soon. What do you guys think about this is open AI going to release a Dali three? I don’t know. That’s pretty crazy to hear. He might just be speaking about this saying like, oh, Dali three is obviously going to come out. Maybe he didn’t really think about that when he said it. But yeah, I made a whole video saying why I don’t think Dali three is coming out. And he definitely is saying otherwise in this video. So who knows? Let me know what you guys think of the comments. Dali three maybe? I don’t know. I would obviously love to see it. But we don’t really know. Anyways, this is a smaller piece of news hugging face, which is known as this website that hosts a lot of these little AI open source projects and such where you can test these demos out for free, been featured on the channel many times for their free demos. They have compiled everything into machine learning tasks. So here you can find what you need to get started with demos, use cases, models, data sets and more in terms of hugging face tasks compiled all of their different stuff into these different little models and segmentations, which is a really nice and organized way of doing things. And I thought it was important for you viewers to know about. So I’ll link that down below if you guys want to take a quick look. This is especially relevant to developers, I think. Which there’s a surprising amount of them on my channel. So thank you for watching all you devs out there interested in machine learning. Similar to Nishi Journey, we’ve got this full anime model known as anything V3. It’s completely free to use in this demo on hugging face, but the wait times are super super long. And as you can see, some sample images from anything V3 here on Twitter definitely show off its capabilities as an anime based of a model. It’s got high resolution, high coherency and it definitely gets the job done, although I do believe these are upscaled. For those interested in anime art, it’s definitely a very capable model. And it’s definitely free for you guys to check out and use with the link in the description anything V3. Apparently if you duplicate the space on hugging face, you can just use it immediately. Alright, now getting into a pretty interesting little piece of AI news. So basically this is a part of scenario dot gg, which is this new AI that’s going to be used to create video game assets. Basically they’re just showing off an absolute ton of different little mini worlds they created with scenario dot gg. Some bigger examples as well. They look like they’re coming out pretty cool. I don’t know that much behind what’s going on under the hood of this model, but it’s really interesting to see some websites starting to sprout up that are using AI to create stuff like video game assets. And here’s the website as a whole as you can see they’re doing a lot more than just many worlds. They’re doing characters and items and stuff like that as well. So very, very cool to see build your own AI engine unlock lots of creativity speed up production chooses a set of visuals trained bespoke AI models and then generate outstanding assets. I’ll probably make a full video on this once it is released. But this is scenario dot gg video game assets generated with AI. And for our final little piece of AI news today, this website I kept getting asked about I believe the creator of this website emailed me or DMed me or something along those lines. This is a huge huge list of just a ton of different a eyes all nice and categorized on one wedge website known as future pdf the largest AI tools directory updated daily. I’ll probably make a full video on this exploring my favorite tools when I get a little bit more time to really take a look at this, but it looks to be pretty darn sweet at five new tools every single day. All right, you can see what is known as popular on this website. And yeah, if you just want to see some of the latest new stuff you might not have known about galactic is already on here mid journey. Obviously copilot stuff like that’s here, but there’s stuff I haven’t heard about before already that I’m seeing hundreds of AI tools. So yeah, this future pdf website looks like it’s going to be huge for finding the latest and greatest with AI and testing these out getting direct links and stuff like that all nice and categorized. So yeah, I’m going to have to take a deep dive into this one for sure and get back to you guys with a full video on it. But I wanted to mention in this video looks really, really interesting. Thank you guys very much for watching this video. I’ve got more to come this week in terms of AI and different projects and stuff like that that I want to check out. Thank you so much to everyone for supporting the channel and watching my videos and make sure you join the discord because everything you might love about this channel is enhanced tenfold on the discord. It is a really, really great little community that we’re building over there and I highly suggest you guys come and contribute to the discord. Thanks so much for watching. I’m at vid pro and I’ll see you next time. Goodbye.