I remember using DALL-E 2 for the first time last year and was utterly shocked. They had such a heavy flow of traffic I had to stay up late just so I could give it a try.
But I remember that when I finally got my hands on it, I was really disappointed with its performance. The image was distorted, inconsistent, and didn’t even make any sense.
I saw the future of AI art but knew that reality didn't exist yet.
Somehow, in a blink of an eye, everyone just began to create artwork and images using AI. The turning point? Well it was Midjourney.
It’s hard not to see its impact in the digital landscape and its potential to disrupt the art industry. But who is behind Midjourney and can you trust the company?
What Is Midjourney?
Midjourney is a generative AI model that transforms text into images. To achieve this, this model combines techniques from computer vision and natural language processing to create stunning artwork.
A major differentiator that makes it different from DALL-E and Firefly is because it’s still in open beta, Midjourney doesn’t exist as a separate platform or standalone website. It's currently just a bot integrated within Discord.
Its gone through five iterations since its initial release, with the latest being V5.2. There has been no solid information yet regarding V6’s release, but rumors point to late 2023 (I'm getting doubtful though).
Is Midjourney Reliable?
Who Is Behind Midjourney?
Midjourney was founded by David Holz, one of the co-founders of Leap Motion. Their greatest goal is to “expand the imagination” of the human species.
The company is currently in the middle of a user surge, with more than 15 million lifetime users and a daily intake of 100,000+.
In the past, I’ve always pointed out that Midjourney isn’t suited for creating realistic images, that they tend to stylize their outputs. Want to know why?
According to David Holz, the main goal of Midjourney isn’t to create photorealistic images that can pass as photographs, but to emphasize the aesthetics as described by the prompt. This is honestly a good and ethical policy to have when it comes to generative AI.
He also said that he's uncomfortable with the “uncanny valley” and specifically avoided them in creating the model.
The Quality of Midjourney Outputs
To measure how good Midjourney is, we need to separate its outputs into categories. This will help us get a better overview of its capabilities and where it falls short.
Like David Holz said, Midjourney actively avoids creating photorealistic images. Despite my prompt starting with “a realistic image,” the generator still created an image that’s pretty easy to clock as AI. The subject’s features are too flawless, has a wax-like texture, and it looks like a still straight out of a movie.
Surrealism is one of Midjourney’s greatest strengths. Even with a simple, one-line prompt, the generator can create stunning images that require a lot of creativity and imagination. The output for this specific prompt is whimsy and unique. It also somehow injected color into paper, something that’s usually lifeless and monochrome.
Even for people like us, it’s hard to imagine abstract concepts. It requires a lot of creativity and nuance. Midjourney was able to do so in just a few seconds. The artwork it generated has a “renaissance” feel to it, which helps evoke feelings of softness and kindness — something that’s present in sympathy.
Midjourney isn’t as good as DALL-E 3 when it comes to understanding context, but it does hold up pretty well. Despite the long prompt, it was still able to give a unique take into what I asked. The artwork successfully provokes a feeling of passing, melancholy, and wonder.
We’ve also done comparisons of Midjourney against other AI image generators. If you want to see more of what it can do, be sure to check our articles out!
What Else Can You Do With Midjourney?
If you ever find an art somewhere that you want to imitate using Midjourney, you can use the Describe command to help you out. This feature lets you upload an image into Discord and its model outputs four prompt variations that closely match your original input.
But how close exactly are the new prompts’ output to the original image? Close enough. For instance:
It’s not exactly perfect, but it captures the intent of the original image. I’d say this feature is best suited for finding specific art styles rather than how to describe the original image. You can see that in the images above, as they’re both imitating oil paintings and a rainy atmosphere but neither subjects look alike.
You can also upload two images into Midjourney and transform them into one using the Blend feature. To give you an example, let’s use these images:
As you can see, Midjourney slightly tweaks some elements of both images to ensure that the blended image has a coherent theme. The old man’s clothes have changed, the layout of the street is different, and the characters in the alley are more prominent in the new image.
This is a great example, but these images are still pretty similar to each other. Both have elements of realism, which makes blending them far easier. So, what happens when you mix two wildly different images?
What Midjourney does is create a middle-ground. It takes the most prominent elements of each one and discards the rest. For the left image, it’s the woman’s hair and the light peeking through the window. For the right one, it’s the art style and the sun. The result is pretty good but, in doing so, I feel like the original intent of each image got lost in the mix.
This is one of my favorite Midjourney features so far. Have you ever generated an image which is perfect apart from one single detail? Usually, what happens is you pick a different variation of the image or generate an entirely new batch. You don’t have to do that with Midjourney.
With region variations, you can pick the areas of your generated image that need to be changed. No need for trial and error, no need for repeating the same prompt over and over again. Just a simple select tool and you’ll get four different variations of that specific region.
Can You Monetize Midjourney Art?
Yes, you can monetize Midjourney art. The company specifies that you own the artwork, but not the copyright. What this basically means is that the image is technically yours and you can use it however you see fit — but so does everyone else. If you find someone else using your AI-generated art, you don’t have the power to claim intellectual property rights.
So, how can you make money using Midjourney? I have a couple ideas in my mind, specifically:
- Sell your art as stock photos. Shutterstock and iStock lets you sell AI-generated art on their platform. From there, you can receive royalties each time someone downloads your image.
- Sell your art directly to clients. You can also sell your Midjourney art directly to clients, either through your own website or through social media platforms. However, you need to be transparent if you’re doing this.
- Sell digital art prints. You can sell digital art prints on your own website, on Etsy, or on other online marketplaces.
- Sell your prompts. There are also platforms like PromptBase that let you sell creative prompts for $2 to $1000.
The Controversies of Midjourney
The root of most Midjourney controversies essentially boils down to creativity. To be more specific, how ethical is it to use AI art even if it’s using your own idea?
Last year, The Atlantic was put under scrutiny after they used Midjourney to create images of the controversial figure Alex Jones. Later that year, Théâtre D'opéra Spatial won the 2022 digital art competition of the Colorado State Fair. The problem? It was created by Midjourney.
There have been lots of debates about whether or not you own artwork from Midjourney considering that it’s trained using millions of images online. Some people also feel that AI art is taking jobs from artists.
That said, the biggest implication of Midjourney and other AI image generators is not yet here. We’ve seen glimpses of what deepfakes can do after the fake viral image of Pope Francis and the people who fell for the Mr. Beast scam, but that’s nothing compared to what’s about to come.
For the most part, we can still spot deepfakes. The mouth movement is unnatural, the tone of voice is monotone, the rest of the face remains still even in emotive situations. But what happens when we get rid of those issues? What happens when deepfakes can imitate real life with an alarming level of accuracy? Only time will tell.
How Different Is Midjourney From Human Art?
AI images often have telltale signs about its origin. For example:
- Subpar text generation.
- Aesthetics over realism.
- Inconsistency in design.
- Extra fingers or non-circular pupils.
- Off-center subjects, often vertically.
- Rendering issues, particularly in the edges.
Do you want to see how well you can spot AI images? Take the test here.
When Should You Use Midjourney?
Midjourney is best for creating artwork, instead of realistic images. If you’re looking for good AI image generators for realism, I suggest using DALL-E 3 or Adobe Firefly instead. You can read more about their differences in this article.
How Much Is Midjourney?
Midjourney has four pricing options. These are the following:
Fast GPU Time
Relax GPU Time
Purchase Extra GPU Time
Work Solo In Your Direct Messages
Maximum Concurrent Jobs
12 Fast Jobs
3 Relaxed Jobs 10 Queued
12 Fast Jobs
3 Relaxed Jobs 10 Queued
DALL-E is OpenAI’s image generator, released even before ChatGPT. Currently on its third iteration, it’s now more powerful than ever before, even showing signs of better prompt understanding and creativity.
Unlike Midjourney, DALL-E 3 can use conversations instead of prompts to create images. This is possible through the robust natural language processing (NLP) of GPT-4, which allows conversion of conversations into a language that the model can understand.
As for the output comparison, here’s what Midjourney looks like against DALL-E 3 with the same prompt:
Both stunning but for different reasons. Midjourney went to a more abstract, almost Lovecraftian route while DALL-E 3 created recognizable elements of the universe. I’ve also already tested these tools in the past and I’ve found that Midjourney is best for stylized outputs while DALL-E is better suited for longer and more specific prompts.
DALL-E 3 can be used with a subscription to GPT-4, which costs $20, or using Bing Image Creator, which has 25 free credits daily.
Adobe Firefly 2
Adobe Firefly 2 is an AI image generator made to fit seamlessly into the Adobe ecosystem. It’s created to aid designers in creating digital artwork by providing free-to-use assets using artificial intelligence.
Across all my tests, Midjourney has always been the better model for creativity but Adobe Firefly is simply amazing at realism. So, if you’re looking for something that can pass as an actual photograph or a stock photo, use Firefly 2.
Unlike Midjourney, Adobe Firefly 2 is free, as long as you don’t mind the small watermark on the bottom left side of each image.
Stable Diffusion is an open source text-to-image generation model that can create stunning images from simple prompts. This means you're free to use, modify, or distribute their code and outputs legally.
I might just be getting tired of Midjourney, but I actually prefer the art style of Stable Diffusion’s image. It’s way simpler, but it conveys a sense of wonder and imagination whereas the former evokes awe. It’s a different but more grounded perspective, which I really like.
Stable Diffusion’s cost depends on where you’re using it. Dream Studio comes free with 25 credits. If you need more, you need $10 for 1,000 credits. On the other hand, Clipdrop lets you create 400 low-quality and watermarked images daily. It costs $9 a month for 1,500 images a day and without a watermark
In A Nutshell
So, is Midjourney reliable?
As someone who’s been using Midjourney for a long time now, my answer is an emphatic yes.
It can create surrealistic, abstract, and even high-context images even with a simple, one-line prompt. As for its other features, it also has a pretty reliable describe command and blend feature.
Other AI image generators may have a specialty, but Midjourney is the most consistent of the bunch. However, if you ask me what I dislike about it, I’d say that it’s really hard to get what you want without persistence. It doesn’t have the nuance of DALL-E 3 yet, so I’ve been doing a lot of prompt experimentation just to get what I want. But, when I do, it’s always worth it.
Above all else, it’s a proof that your ideas are worth something. After all, Midjourney art can be monetized since you own the artwork. As we shift towards a future that’s more dependent on AI, it’s best to invest and capitalize in Midjourney now.