LARRUZ demos OpenAI's new image generation and editing tools in ChatGPT. The host walks through 10 use cases: product marketing images, photo retouching, character-consistent comics, meme remakes, textbook visuals, and research papers. The model generates images without mimicking specific artists' styles.
OpenAI just came out with their new image generator AI within ChatGPT and it is so good it took my breath away. I guarantee that you'll see things here you've never seen before. Why is that? Well, we are going to look at 10 incredible examples. 1. You will not believe this. It can create a newly imagined Apple style product from the series called Severance. So how authentic is this? Would Apple like this whole? Well, hold on to your papers fellow scholars, because this is an image from the official Apple website. They put it up there. I am not kidding. The link is in the video description. Hope they will still have it up by the time you're watching this. And yes, with that said too, I think it can do any style in marketing images. And my goodness, these are so good. Some of these are ready to go as they are. What is happening? But it gets better. 3. It also has image editing capabilities. Reimagine it however you want. Kind of like Photoshop, only better. How? Well, this is an image of me having the honor of talking to four legendary scientists and I had it remade in a different genre. Note that the text is amazing, but it forgot my 2 minute papers business card so there is that. But wait. I explained this mistake to the AI and then I got this. It fixed it. I fell off the chair when I saw this one. Wow. Unfortunately it cost us the glasses of Jim Fan. I am hoping he is okay with it. And it gets even better. I realized how tiny I am in these AI made images, but not quite in reality. I asked the AI to make me regular size and ha. It made me even tinier. Point taken little AI. But Jim got his glasses back so there is that too. 4. And if that somehow is not enough, you can even reimagine yourself with other fellow scholars and put yourself in different situations. 5. Before you ask about redoing memes, it is of course excellent at that too. Especially if you let your imagination run wild. 6. Character consistency is second to none. Can finally really get the same character in different situations. It won't forget. But wait, that means that creating AI generated comics is finally possible where the AI is there to help out. But the idea is coming from you. This looks amazing. Now, while we are looking at some beautiful results here, I would like to note that I am showcasing results that are not in anyone's particular style. You see it in the news that everyone is making Studio Ghibli style images, but. But we won't do that here. That means that I had to discard like 99% of the results available online and generate many demonstrations on my own. So this video took a bit longer. Hope you fellow scholars appreciate it. 7. They showcased a cherry picked image for text generation which came out amazing. But this is the best of eight images so I am wondering does that really work in practice? Well I tried it myself and yep, image generation for me seems to be best in class by a mile. A huge step forward. And wow. Not just for the text being closer than ever to perfect here, but wait a minute, are you seeing what I am seeing? It is not just another regular AI model creating an image the usual way. You see that it already has some high level structure planned out in advance. It seems fundamentally different from already existing systems. So cool. Eight textbook style explainer images. One thing I really missed from previous AI systems was that none of them could do that but this one. Wow. Absolutely nailed it. And you know what the best part is? It also has a really intelligent AI behind this. So if you ask about obscure things like light simulation algorithms that almost no one knows about, not a problem. This is insanity. Nine. Of course, nerve papers. Wanna see how the research papers of the future should look like? Certainly not like this. More like this. Oooh. Loving it. 10. I kept these for last. These are some of my favorites. Dear fellow scholars, this is two minute papers with Dr. Zsolnai. I cannot stop having it redo memorable images of me and my daughter. Now there is one problem here. I would like to look a little more muscular. Oh yes, that's more like it. And I gotta say some of these are super funny while others brought tears into my eyes. Wow. This was an unforgettable experience. Thank you so much for these and onwards to many more. So fellow scholars, this is the age of AI and from now on imagination will become the ultimate tool. What a time to be alive. And once again I made sure that none of these images mimic any one person's style. So what do you think? What would you fellow scholars use this for? Let me know in the comments below. We need new tools for the era of LLMs and weights and biases. Now has weave a lightweight toolkit to confidently iterate on LLM applications. Use traces to debug how data flows through each step of your app and use evaluations to measure your progress. It is the best. Try it out now at WNB Me Papers or click the link in the description below.
