GPT-4o Image Generation Legality: Navigating Copyright Concerns with Studio Ghibli AI Art

Craig: so a couple of days ago, openai released gpt four o and image generation and a bunch of new features in. and wednesday, i was doing some image generation, and i was shocked at how good it was at rendering text and taking i put my logo in, and it created all the stuff really matching the style and images and whatnot. it created great images. unfortunately, i had an event that day, and it took me all day. and by the time i got back to look at it on thursday, they had pulled a lot of their image gen capability off the site. it's back on, but it's not nearly performing as well as it should be because, apparently, we're melting down chips. what's happening here, john?

Jon: yeah. so this is cool. so, yeah. so the the cool thing about this new gen is the text that it can do. so, for those of you who've not used image generation tools, one of the biggest struggles that image generation tools have is with text and just and, proper text. not just writing it properly, but displaying it properly so it's clear, it's concise. and this, when it was released, was a game changer. it was high def text. it was written properly. it was awesome. so but word got out fast, and so due to overwhelming demand. and the other thing too that they were concerned about is copyright concerns. so they pulled the free version, just one day after its launch. so this i guess, what this shows from my perspective is that these companies are still facing challenges balancing user demand with legal, technical constraints, all that sort of stuff. so i'm curious as to and yeah. so the the what you're saying about melting gpus, i i did a deep dive into that. and so, like, he was speaking metaphorically, but still, like, it was overheating their gpus. it was just it was straining them ridiculously crazy. so, it's i like it. this is awesome. that they that they've done, this huge leap. even grok, which i liked grok image generating tool, it couldn't handle text very well. but, but, yeah, what are your thoughts about this going forward, craig?

Craig: for me, this is a huge leap forward. and i will just give you some examples here. these aren't necessarily the greatest examples from quality image generation, but it created this from scratch with the text all by itself. i didn't have to do anything. and i'm gonna share a couple of more just because, you know, i have a few examples. it, like, it was amazing, and i was quite shocked at it. i've only seen, like, ideogram, do this as well, and ideogram's fairly pricey when it comes to creating graphics and whatnot. so it created these with all the images. so it got very popular not only for how much text it was able to create, but the fact that you could throw in copyrighted images or whatnot and say create something in this style, and everyone was going nuts with it. i really enjoyed that because i took our company logo that i created and said, hey. let's create, graphics that have this style, and it went ahead and did all this with the text and whatnot. and it looks great. so that by itself was amazing, but it was, shocking to find out that, it was pulled down for free users. right? next day, i'm i'm a paid user, so i still have access to it. when was creating images today, it was failing a lot.

Jon: it?

Craig: it wouldn't quite create it. i think it's still, having problems with the cpu. i still ended up producing, probably about eight good images today from it when i probably should have produced 15 or so, just from the failures and whatnot. so it's hit and miss still just because i think the performance is not there, but the graphics are great, and people are loving it. and i think as soon as they tweak the performance a little bit and figure out how to, let, people actually create what they want without, overheating their servers as it were, this will be an incredible tool. when i saw this, i immediately texted a friend of mine, alex, as you know, and said, bye bye, canva. this is the end for them.

Jon: yeah. well, i use, like, for, like, i yeah. canva's canvas is good. when it comes to presentations, i use gamma.app. we're not sponsored by them. just i find so they do presentations, and it generates the slides and does all the like, from the content that you give it, it can also generate ai content. so, it's cool. i think obviously, once they work this out and work this further, that will overtake that as well or just enhance it. but, i think i think this is crazy. this is one of those, i feel, breakthroughs in imagegen that, that's been needed for a while because although they can generate decent images, the text is not good. and i'm i'm curious to see what other features, are included in kind of this advancement, if that makes sense.

Craig: yeah. gtv four utilizes complex algorithms to reproduce styles like the ghibli, one that's been going around everywhere. it's auto, autoregressive image generation combines the analysis of diverse of datasets with user defined prompts. so this gives a key advantages, improves, character proportions and, scene clarity, higher fidelity and style recognition, and consistent output across, various images. so there's a lot of copyright debate when you can just put something in. a lot of, image crashers won't do what this is now. i haven't asked it to, create a image of elon musk, for example, or, like, my go to is, elon musk and, mark zuckerberg jousting on dolphins and see if they can create that image if, like, if they could do that and create a good image, that's probably a pretty good image generator. the problem is most of them, as soon as you say a person's name or a company's logo or a brand, they're like, ah, we can't do that. like, yeah. there's a reason why a lot of these things i don't know if you noticed that. if you asked it to do a picture of a rose, they would refuse. really? and then you if you told to do a picture of, the flower rose, no problem. so there's something to do with rose that is a trademark or a name that it's inferring. it's saying, yeah. i can't do that. and so a lot of times when our servers are too busy or there's a there was a problem generating your image, come back later, a lot of times that's just copyright issue. they just don't have a good error message for that and don't know how to tell you. because you know, it's just what it is. because you put it in a, image generation. it says, yeah. we can't create that image. ah, that's why i'm getting these errors. you start to feel about that. so openai asserts that it doesn't replicate individual artists, but instead focuses on a broader studio styles. this claims remains controversial, particularly regarding unique influencers like hayao miyake's distinct style miz miyazaki's. sorry. potentially devaluation of their work, loss of income opportunities, and attributed use of their artistic styles are some of the concerns from the artist. so we'll have to see, what's gonna happen with the legal copyright cases that are going on. yeah. because in the end, i think it will affect how good some of these, image models are because i believe they're gonna have to remove some stuff or remove some features or, not allow people to copy styles, whatever it is. text rendering clarity is superior compared to competitors. multi prompt handling is more effective at processing elaborate prompts, and you can tell it stuff like take this out, take that out. so it's not too bad. it's consistent across sequences, so it maintains coherence and narrative and character design. and style fidelity captures essential aesthetic qualities of this studio ghibli. but, this is great. like, i literally, it creates, use your colors, creates it in your style. branding becomes easy. i'm thinking more of, like, from a social media branding company point of view. you can now create art and images for your website that have the colors, the branding, and everything automatically that's consistent with the fonts. it's amazing, and this is going to be a game changer once they get the performance working the way they want.

Jon: absolutely. yeah. i'm excited. i'm excited for this. yep.

Craig: so let us know what you think. are you using these tools to generate art? i know there's a lot of people out there who are anti that, and some people came at me online for some crappy images about whatever that i created with adar. they were actually not great. but it's you know, they need to be going after these billion dollar companies if they want anything to change. and because, apparently, ai generated images are not art, and yeah. people will tell you that right away. so what do you think? is ai generated images art? what would take it across the threshold? are these image generation generators gonna be neutered by copyright laws that, in favor of the artist, or is it gonna the artist's gonna be run over? we'd love to hear what your thoughts are on this and the future of ai art and copyright. absolutely.

Jon: yeah.