OpenAI’s DALL-E 2 - AI-Based Art Is Here! 🧑‍🎨

Dear fellow scholars, this is two minute papers with Dr. These amazing images were all made by an AI called Dall? E 2. This AI is endowed with a diffusion based model, which means that when we ask it something, it starts out from noise, and over time it iteratively refines this image to match our description better. And over time, magically, an absolutely incredible image emerges. This is a neural network that was given a ton of images and a piece of text that says what is in this image? That is one image. Caption Pair Dall? E 2 is given millions and millions of these pairs. So what can it do with all this knowledge? The key takeaway here is that it does no or very little copying from this training data, but it truly comes up with novel images. How? Well, after it has seen a bunch of images of koalas and separately a bunch of images of motorcycles, it starts to understand the concept of both and it will be able to combine the two together in into a completely new image. Previously, we also compared these works to what real human artists are capable of. And in my opinion, this tool should not be framed as team humanity versus Team AI. I believe this should be framed as team humanity supercharged by team AI. Here is an example of that. And the results are so good and it is so easy to use. I said on several occasions that this is going to change the world of art creation as we know it. And you are an experienced fellow scholar, so you may ask the question. Okay, but how? So here are five beautiful examples of how artists are already using it. 1. Texture synthesis. Oh yes. Whenever we see these beautiful animation movies or video games, we see a lot of real looking content in there. The roads are realistic, the buildings and hills are also realistic. We can get a great deal of help in making them this realistic by taking photographs of real objects and importing them into our virtual world. But there is a problem. What is the problem? Well, let's see together. Let's take a photo. There is a stain on the wall. Is that the problem? No, that's not the problem. We can cut that out. Now we need to extend this texture. But since this is just a small snippet and we have long walls and big buildings to fill in a video game, so what do we do? Well, we start tiling it and. Oh, now that is the problem. It is quite visible. What happened to this texture? It has been extended, but at the cost of seams and discontinuities appearing in the image. So how do we get a seamless texture from this? Well, Dall? E2, hopefully understands that this is a rock wall and. And can it fix this? Now hold on to your papers and let's see together. Yes, it can. Amazing. Now let's have a look at it attached to a real virtual object and see how it looks in practice. Oh my. This is gorgeous. I love it. Here is an even more difficult example with paint peeling off the wall. This is the real world phone photo and the tiled version with the seams. And now Dall? E 2. Wow. I could never tell that this image is not a real photo. And have a look at it added to a video game. So good. By the way, these are Stan Brown's works who used Dall E2 to great effect here. Make sure to check out his work. For instance, he also made these amazing materials for for one of my favorite games of all time, Path of Exile. The link to his work is available in the video description. And just imagine using the free 3D modeling tool Blender and this plugin to create roads. And you can make it an asphalt road, dirt road, or make it look like pavement in just a couple clicks. With Dall E2.2, Dall? E2 is now able to generate real human faces too. And sometimes they come out really well and sometimes not so much. No matter. In these cases, we can use a face restoration program to fix some of the bigger flaws and run it through a super resolution program to add more details. And bam. There we go. Fantastic. And if we have a portrait that works for us, we can plug it into this amazing tool called Deep Nostalgia to create a virtual animated version of this person. How cool is that? What a time to be alive. 3. This one speaks for itself. The Cosmopolitan magazine made their first cover from an image that was made by an AI. No tricks, no frills, just a good old image straight out of Dall E2. And the kicker is that some of them are already good enough to become magazine covers. That is incredible. Talk about the age of AI image creation. 4. This tool can even help us create new product designs. And I mean not this one. Although this toilet car might have some utility if you need to go when you're on the go. But I meant this one instead. And finally, five, we can also combine Dall E2 with other tools that take a 2D image and try to build an understanding as if it were a 3D scene. And the results are truly stunning. We can get so much more out of these images. And one of the important lessons here is that not only AI generated art is improving at an incredible pace, but an additional lesson is that everything is connected. So many of these tools can be made even more powerful if we use our imagination and combine them correctly. This is huge. For instance, Nvidia is already building virtual copies of the real world around us. A copy of the real world? Why is that? Well, because this way we can reenact real situations for their self driving cars and even create arbitrarily complex new situations. And these AIs will be able to learn from them in a safe environment. Self driving can also be combined with a number of other technologies that make it an even more amazing experience. See the assistant there? It sees and identifies the passenger and understands natural language. It also understands which building is which and what shows are played in them today. So we might jump into the car not even knowing where we wish to go, and it would not only help us find a good show, but it would also drive us there. And plus one, remember the shoe design example? How about trying on pieces of clothing, but not one by one as we do in real life, but trying on all of the clothes at the same time? Yes, these are all AI generated images. And it is not just one still image, but video footage where the camera moves around. This is spectacular. Imagine what buying clothes will look like when we have algorithms like this. Wow. Or if an actor did not have the appropriate attire for a scene in a movie. Not to worry, just change it in post with the AI. How cool is that? You see? Once again, everything is connected. So does this get your mind going? What would you Combine with Dall E2 if you're looking for inexpensive cloud GPUs for AI? Lambda now offers the best prices in the world for GPU cloud compute. No commitments or negotiation required. Just sign up and launch an instance. And hold on to your papers. Because with Lambda GPU Cloud you can get on demand a 100 instances for $1.10 per hour versus $4.10 per hour with AWS. That's 73% savings. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Thanks for watching and for your generous support and I'll see you next time.