Summary
- GPT-4o creates stunning, realistic images more efficiently than Midjourney.
- ChatGPT’s multimodal AI can intelligently regenerate old Midjourney pictures, enhancing image generation.
- ChatGPT’s integrated chatbot and image generation tools revolutionize the collaborative image creation process.
I’ve been a loyal fan of Midjourney for a few years now, especially after the V4 model dropped and blew everything away. At least in my opinion! Now, however, I’m switching teams to ChatGPT’s latest model, which replaces DALL-E.
I don’t think I’ve ever seen just a dramatic leap in ability and quality in one go, and the more I’ve been using GPT-4o’s new image generation feature, the less I’ve felt the need to use Midjourney. Now, that subscription is canceled.
GPT-4o Creates Stunning Images
Let’s get the first and most important reason I’ve switched out of the way—the images look absolutely amazing. Midjourney has always excelled at creating nice artistic images that look like paintings, or digital art. However, I’ve always felt it falls a little short when it comes to more utilitarian images. The sort of thing I need to create when none of the stock image sites I use have exactly the right image.
Now, GPT-4o can make realistic, utilitarian images that look photo-realistic without much fuss. Midjourney still has a particular vibe that prevents these sorts of images from looking real, but GPT-4o has shockingly convincing results.
It got our logo wrong, but according to the bot itself, this is by design to avoid trademark infringement. Either way, that’s not a real person, and not a real shirt, and honestly I wouldn’t have known.
Add to that, GPT-4o is now also capable of similar artistic flair to Midjourney (such as it is), and it seems smarter to just go with 4o.
It’s not just because of pure image generation abilities, but because my subscription to ChatGPT includes everything else I use AI chatbots for. So it’s a net reduction in subscription costs. I’d been holding on to Midjourney, because ChatGPT could not generate images I considered usable, but that’s no longer the case.
Related
I Can Actually Fix My Old Midjourney Pictures
One of the best things about image generation with ChatGPT is that the multimodal ChatGPT AI can look at images you provide and intelligently use them with its own image generation system.
For example, here I took a behind-the-scenes photo of Patrick Stewart as Captain Picard with hair, and asked ChatGPT to change it to a mullet.
So you can see, although the new image is completely generated from scratch, all the same basic elements have been replicated. This means I can take Midjourney images that I can’t get right, or I’m not happy with, and feed them into GPT-4o, asking it to fix it. Like this image that’s supposed to be Neo from The Matrix, but Midjourney kept getting the face wrong.
What’s really cool is that ChatGPT first analyzed the image to identify what it thinks isn’t working, then had a quick back and forth with me to nail down what I agreed with and what I wanted.
Image Iteration Is a Game-Changer
It’s this iterative, back-and-forth conversation that really makes me open ChatGPT instead of Midjourney when I want to make something. Midjourney is just an image generation model, but using ChatGPT feels more like working with an illustrator or artist, giving them examples of what I want, asking them to make amendments to existing images, and generally collaborating to create the images I need.
I have never felt like I was the “artist” when using image generation software, but more like I was still just a client commissioning an image. Except in this case, the service is being provided by artificial intelligence. The only problem is that the software is uncommunicative, and I have to hope that my prompt rubs it the right way.
Now, with GPT’s chatbot and image generation powers combined, it absolutely feels like I’m engaging an intelligent entity that understands what I’m asking for, and can look at its own output and see that it’s messed up when necessary.
GPT’s Knowledge Sets It Apart
GPT has a huge store of general knowledge that let’s it enrich its self-prompting for image generation, understand images that you provide, and understand context when you ask for something. I particularly like when ChatGPT quizzes me on details that I did not think to ask for, so that it has a more complete idea of what I want.
This has resulted in a much higher success rate, and I spend less time waiting for image to generate and more time getting exactly what I wanted.
I’m Still Discovering New Tricks
Because the GPT-4o image generation tool is more than the sum of its parts, I’m constantly discovering new things I can do with it. The ability to build up the concept for the image in detail first, in conversation with the bot, means I can now attempt things that simply wouldn’t work in Midjourney. Especially when it comes to layout, prompt adherence, and text generation.
I am still doing a lot of the same stuff I was doing with Midjourney, but it’s just more intuitive, more successful, and now with this new advanced image generation model I’d argue it looks just as good, if not better, than Midjourney in most cases.
That said, as I write this, Midjourney’s V7 model is in the alpha stage of development, and I wouldn’t bet against the company bringing out the big guns to compete with GPT-4o. So who knows, maybe I will switch back again if the new tech impresses me enough.
