OpenAI Image Generator Upgrades (Dalle-E 3)

Chatgpt's abilities are about multiply because we just got an announcement of OpenAI's brand new product, Dall E3, which in early October will come with your ChatGPT plus subscription. And this is a very big deal because it's a major step towards multimodal language models. And my favorite part, this is going to be user friendlier than anything they've ever released before. Essentially, we're getting a midjourney that is really easy to use inside of ChatGPT in just a few weeks with such a major change in their product. There's a lot to talk about here. And at the end of this video, I'll talk about where this entire space is heading, because if it wasn't clear before, this one paints a clear trajectory to where these AI products are heading next. So let's get into this. Okay, so the first point we need to cover here is that we're actually going to get a better mid journey inside of ChatGPT included with the price of ChatGPT Plus. That's a pretty amazing offering, especially considering the fact that it's actually objectively better. Now, I say that for a few reasons. The first one would be that the generations are on par with Midjourney. The other artistic outputs are unbelievably good. As good as Midjourney. The photo listen really is there. As you can see, they also pretty much figured out the hands. But here's the part that makes this better than Midjourney. As you can see, it can generate text, which up until now was not possible with Midjourney. And there were some other models that did this, but they were kind of terrible at everything else, so it doesn't really count. And the last reason why Dall E3 is better than Midjourney is the fact that it's able to generate consistent characters. So as you can see in this little video that they used to launch it, the key part here that might be overlooked is that as he generates these images, the character stays consistent. Look at the hedgehog in this one and in this one and the next one. It's the same hedgehog, it's the same color palette. You're going to be able to generate stories that are consistent now, which was an incredible pain inside of Midjourney. Very hard to do. Borderline impossible, to be honest, the character always looked a little different. It just couldn't do consistent characters very well. Now we have that. Okay, but beyond the output just being better than Midjourney, what else makes this so darn special? Well, it's a massive leap in terms of user friendliness in AI, because what this is going to come with is a prompt generator. What that means is you're going to type in messages just like you would speak to a human, without crafting your prompts carefully, without introducing certain keywords that will produce desired effects, as the demo suggests. You can just say, my file five year old keeps talking about a super duper sunflower hedgehog. What does it look like? And that's the entire prompt. So in order to get this result inside of my journey, you would need to precisely describe all of this. You would need to include the butterflies, the fact that there's multiple sunflowers on his back, the style, the lighting, all of that. Here you simply say super duper Sunflower Hedgehog and Dall E3 uses GPT4 to generate your prompt for you. Now, this is no new technique. As soon as GPT4 came out, I created a video detailing this exact prompt process, where you can use it as a prompt generator to create your image prompts. But now it's all built in. You don't need a particular prompt to generate your secondary prompt, where you copy the output inside of ChatGPT, then head on over to Discord, enter a channel that has the Midjourney bot enabled, type imagine, and now paste your prompt. No, all you do is describe what you want and you get the result. And honestly, I feel like this is the biggest deal with this announcement. Yes, the image generation is fantastic. Yes, this is going to be super useful for marketers and content creators. But the big innovation here is that we're moving away from a need for prompt engineering where there's no way around it right now, to a universe where you're just going to talk in natural language and it's going to do the prompt engineering for you. Now, this doesn't make prompt engineering useless. If you want specific results tailored to your very own needs, you're going to need to go in and tweak the specifics. And that's why it's a fantastic skillset to have, especially as the capabilities of these models go up. You're always going to be able to go in and fine tune the results for your own needs. But the point is, for the first time, you don't have to. You can just type in something simple, get a result, and run on with it. And frankly, for image generation, where usually the precision of the outputs are not as critical as when you're handling data or extracting some concrete info from a long piece of text, honestly it might not matter. You just want a super duper sunflower hedgehog. You don't want to bother with all the specifics. Just give me a picture of a hedgehog with sunflowers for my five year old. Right, and now it does this. Okay, so what this makes clear, if it hasn't been clear before, is that over time these tools will become more user friendly, they will become easier to use, both in terms of prompting and the way you access them. Right. As of now, this is way simpler to use than Midjourney. And just at this point I want to point out that yes, Midjourney did announce in their most recent office hours they will be offering a web interface in the next version and that version 6 is coming and there's also going to be a text to 3D generator soonish. But from everything that I see, it seems like Dall E3 is going to be available earlier than all this. And honestly, this prompt generating feature is kind of the big selling point here because this will make AI art accessible to the masses. Whereas right now with stable diffusion being very, very niche and midjourney being only accessible for Discord plus, the prompting is not that easy. Well, turns out that not that many people keep returning to it over and over again. Whereas if you have this inside of ChatGPT, it's just going to be super simple. I reckon there's going to be a new thing in here where you're going to get to generate images. Okay, so as mentioned, this is coming in early October and it will be included with the ChatGPT plus membership. And that brings me to the last point here. Where is all the settings? So we talked about how it's better than Midjourney, we talked about how it's easier to access it, but what's the direction we're heading in? Because this is painting a picture that is becoming increasingly clear. So as you can see with the GPT plus subscription, you get three different modes right now, right? You get the default where you just use GPT4, you get advanced data analysis, aka the code interpreter before this, where you get to use a different model that actually creates and executes code locally. Plus this lets you attach multiple files right here. And then we have the plugins, which at this point we're at over a thousand plugins, most of them useless, but for some use cases it's nice to return to them. We talked about plugins often enough on this channel, and then last but definitely least, we got the custom instructions recently, which is extremely useful. I use this about 95% of the time. Again, I have multiple pieces of content on this and this is really the one feature that everybody should be using. So if you're not familiar, check out those videos on my channel. But here's the entire point. They've been adding multiple functionalities step by step, right? Better model. You can attach files, bunch of unique capabilities with plugins, and you can set your own custom profile inside of custom instructions. And now they're adding an image generator right inside of ChatGPT Plus. And if you bring all those ideas together into one model, you arrive at the end points that we're aiming at in the short to midterm, which is a multimodal AI that will allow you to upload an image and start the conversation by saying create something similar. And then you will get an image back or you feed it a blog post and say, now create a visual for this blog post. And it will simply do that. And why I think this is so important is that this shifts so many industries because right now we're at the point where a lot of the AI hype fell off and only select professions rely upon GPT4 for their daily work heavily. But as we move closer to this multimodal future of this technology, that honestly, with this release seems to be right around the corner. You'll see a lot more people using this in their everyday life because very soon, 5 year olds will be able to create their own children's book just like that. And grown ups will be able to upload an Excel sheet and the last yearly report and tell it, hey, please create this year's report from the data that I provided to you. And it will not just do it, but because you set your profile up that way, it will also immediately write the email to your boss for you. And once more and more people experience these advanced capabilities, they won't be able to go back anymore, just because soon it's going to be really simple to get actually useful results. All right, so if you want to learn more about image generation, this video right here includes a massive free resource that teaches you how to prompt manually if you want to do that. And you should definitely check out our discord. Every Sunday we're holding AI art competitions where soon you're going to be able to participate with Dall E Free. I'll see you there.