22 hours Ago
Both Midjourney and ChatGPT have recently released new versions of their AI image generators. Historically, these have been two of the best options out there, pioneering the space for what has come.But, when put against each other, which is best? Midjourney V7 or ChatGPT 4o image generation?I put ChatGPT vs Midjourney to the test using seven different prompts to see which is the best AI image generator.
These test everything from the model’s ability to understand context, recreate complex shapes and think creatively to make images.ChatGPT vs Midjourney V7: The rulesWhile both models create images, it can be hard to make this a fair matchup, mostly because of the amount of settings Midjourney allows you to change. With that in mind, these were the steps I took first.
For Midjourney, I used version 7. This is the latest version but it is still in an experimental phase. I also tried the prompt with both personalization on and off (the setting that adds your preferred art styles to images).
Midjourney produces four versions of each image compared to ChatGPT’s one attempt. In all cases, I chose the best image from Midjourney and upscaled it (asked for a higher-quality version).1.
PhotorealismLeft: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Create a photorealistic image of a puffin flying over a cliff face with water below. In the background is a mountain range. It's a sunny day and below the puffin is two people looking at it through binocularsChatGPTThis hits almost all of the marks.
The image, while potentially over-saturated, is photorealistic. There is a puffin flying over a cliff face, there is water below and there is a mountain range in the background.On top of those points, it included the two people looking through the binoculars.
Sure, they aren’t looking at the puffin but otherwise, this is pretty spot on.MidjourneyThere is a lot going on here. I can’t disagree that everything has been included.
Mountains in the background, a puffin, two people with binoculars, and even water and mountains in the background.However, let’s address the elephant (or puffin in this case) in the room. The puffin is giant and could take on Godzilla if needed.
The image also isn’t really photorealistic, looking slightly more like an oil painting than anything.Even with puffin sizing issues aside, I still think ChatGPT understood the cues more accurately. Both models created water underneath a cliff face, but ChatGPT understood the context of the prompt more accurately.
Winner: ChatGPT wins this one in just about every single way. While I'd love for puffins to be giant mythical creatures, Midjourney just misunderstood way too much context here. ChatGPT, on the other hand, nailed the brief.
2. Complicated promptsLeft: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: A large market with a stall selling fruit, one selling dresses, and one selling ceramics. In the background is a river, and in the far distance is a forest.
A man hands a woman money in front of one of the stalls, and two kids are running through the middle. In the sky is a hot air balloonChatGPTA lot was going on in this prompt, and it could be easy for an AI model to ignore some of it. However, all of the key details are here.
The hot air balloon, the two kids running through the middle, and the man handing a woman money. It is also, clearly, a hot day, and you can see the market selling fruit, ceramics, and dresses.Despite all the details required, ChatGPT produced a high-quality and very detailed image.
MidjourneyWhile Midjourney achieved the same image, it was the smaller details that were off. When zooming in, faces aren't complete, the two people's hands are morphed together, and most of the background is a blur.Winner: ChatGPT takes this one.
While both look correct at a glance, and mostly get all of the features asked for, Midjourney is just missing too many of the finer details.3. Adapting real imagesLeft: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Turn this image into a Renaissance portraitThe original image given to the AI models (Image credit: Future)ChatGPTWith this prompt, ChatGPT essentially turned my image into the stylings of the Mona Lisa.
Again, I can’t really fault the model’s work here. It put the exact photo I supplied into the stylings of the Renaissance era.It also does a good job of keeping all the features like headphones, the background, and the clothes I am wearing, while keeping to the theme.
MidjourneyYes, this was the best of the four attempts Midjourney gave me. I do see where the model was trying to go here. It just couldn’t quite make it.
I even tried altering the prompt slightly to make it clear I wanted it in the style of a painting and that made things worse. I assume the brown border is also supposed to fit the theme? It's hard to tell.Winner: ChatGPT has proved that this new model thrives when it comes to putting a creative twist on your own images, and this is more proof of that.
It did exactly what I asked for. It seems like Midjourney got halfway through and gave up.4.
Movie postersLeft: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Create an exciting poster for this film: A Cyberpunk movie set in the year 2250. It is set in a big bustling city. The film is about a detective set back in time to stop an upcoming war from happeningChatGPTIt’s not the most exciting poster ever, but ChatGPT definitely nailed the brief here.
Our detective takes centre stage, with a bustling (and rather futuristic city) nestled in the background. It did take the prompt quite literally for text, adding the requested data with a slogan.Overall, it’s impressive.
The detective is detailed with a neon light shadow on his back, there’s a flying car in the sky, and, while slightly crudely drawn, lots of futuristic skyscrapers.MidjourneyWhat Midjourney lacks in detail here, it makes up for with style. Arguably, the skyscrapers look better here, and there is a lot more to see in this image.
Sadly, Midjourney falls behind with its blurry details. The images on the ground have morphed, the car and motorbike have glitched, and there are lots of bizarre details in the background.While it’s more interesting, there is just too much wrong here.
Winner: ChatGPT did everything I asked for and made a poster that I could put out into the world and no one would bat an eyelid (other than the incredibly boring film title).Midjourney, on the other hand, just got too many things wrong here. I do, however, like the direction it was going in.
5. Text generationLeft: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Make an image of a poster, on the poster it says: "The band AI image generator - playing here tonight at 8pm! Covers of all your favorite hits" Stylise the image as if this is a poster for a band playing at a popular venueChatGPT There’s a bit of a theme with ChatGPT’s image generation. Detail often trumps style.
This poster did everything I asked for, and more importantly, got all of the text exactly right.ChatGPT, just a few months ago would have struggled with this, so it is exciting to see how far it has come.While the poster is boring, it has hit the brief and achieved a tricky challenge for AI models.
MidjourneyI appreciate that Midjourney made this more of a poster at a venue, putting it on the wall outside. I also like the energy it was going for with the picture of the band in the middle.However, other than the words “The band” not a single bit of the text is readable.
Compared to ChatGPT’s ability to get all of the text in its entirety, this feels like a bit of a letdown.Winner: ChatGPT might not have been incredibly interesting here, but it completed the task perfectly. As Midjourney showed, it is not always easy for AI models to deal with text in images.
6. Hands Left: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Make an image of a person's hands, the left one is holding an orange and the other is holding a glass of waterChatGPTAh, how far AI has come. When AI image generation first came about, one of the easiest ways to identify it was hands.
They would have incredibly long fingers, or fingers sticking out of the wrong places.Now, while the hands here don’t quite look completely human, the accuracy is really impressive. Both hands have the correct number of fingers (good start), the water in the glass properly morphs the visual of the hand behind and you can see veins and nails.
MidjourneyMidjourney did a fantastic job here. What I think is especially impressive about this image is the detail. The arms on the hair, the veins, the bruising on the knuckles, and the stretch marks on the hands.
While the ChatGPT image is instantly recognisable as AI, this could pass for someone’s hands. The only noticeable issue is the finger behind the glass not looking quite right. It is also a very strange way to hold an orange, but each to their own.
Winner: Midjourney stole a win on arguably one of the best-known flaws of AI. This goes to show how far it has come. This isn't to say ChatGPT did badly, it just didn't quite match up.
7. Food Left: ChatGPT / Right: Midjourney (Image credit: ChatGPT / Midjourney)Prompt: Make a picture of a bowl of seafood pasta that would be used on a food InstagramChatGPTThis is the kind of food image that I would see in a cookbook and not question for a second. Even though this doesn’t exist, I want to eat it.
Can’t really fault the AI model here; it did everything that it was asked for, even if the random bit of herb in the bottom is very out of place.MidjourneyJust like ChatGPT, Midjourney did an excellent job here. This looks like a real bowl of pasta that you would get in a nice restaurant.
There are even some random tomatoes and garlic scattered around, I assume for decoration.Winner: ChatGPT takes the win here thanks to an ever-so-slightly better image quality, but like the hands, this was close.Verdict: ChatGPT winsSadly for Midjourney, this wasn't even close.
Of course, it is important to note that Midjourney has just released this version and it is still in an experimental stage. However, this latest version of GPT image generation is only a week or two older.While the model's were occasionally evenly matched, ChatGPT just so often excelled where Midjourney didn't.
I do hope Midjourney sees improvements through the testing phases of version 7 as it could be such a great AI image generator.More from Tom's GuideYou can now use Google's AI to make videos from text — and I'm already obsessedI used Grok's free new Studio AI tool to make a website, research paper and browser game — here's what happenedI'm a personal trainer — ChatGPT built me a 15-minute mobility routine for looser hips, and I’m surprised by the results.
