What is Midjourney?
Midjourney 5 creates the most realistic AI generated images so far, but what is it? Midjourney has become arguably the most popular artificial intelligence (AI) for text-to-image generation.
Its creators released Midjourney 5 this week, which has presented the world with another huge step forward, a topic I’ll cover in this blog post. For those that do not know what Midjourney is or would like more details, my partner, Bill Adams, wrote an article about Midjourney a couple of months ago that gives you the details: AI Generated Art Will Supplant Human Artists
The Basics of Midjourney Prompts
Let’s start by looking at text prompts which are vital for the AI to know what you want it to create. Midjourney 5 is much improved in its ability to understand natural language. When providing prompts to previous versions, it was often better to break what you wanted into keywords that were separated by a comma. For example, you might write a prompt such as “a soccer player, soccer stadium in background, crowd cheering, soccer ball, goal, green grass, banners.”
The new AI, however, actually performs better with natural language prompts. So, using our previous example we would type something like “Show a soccer player kicking a soccer ball on green grass with a crowd in the stands the background.” In fact Midjourney Studios, the creator of Midjourney AI, tested these two types of prompts countless times. They say users get far better results from the new version by using natural language. Now that we understand how prompts have improved, let’s see the latest version in action!
Examples of Midjourney 5 in Action
For my first prompt I wanted to see something simple. I used the prompt “a man sitting next to a dog on a cliff looking out over the ocean watching a sunset.” As you can see, I got exactly what I asked for, and the quality is so much better and more realistic than Midjourney 4.
Midjourney 5 Image of Man Playing Frisbee with Dog
Next I typed the prompt “a man playing frisbee with his dog on a beach with waves crashing in on a clear day.” Again, it provided exactly what I wanted to see, and the quality is also very good.
Midjourney 5 Image of a Couple on a Boat Watching a Sunset
I decided to try something a little harder by prompting ” a couple on a boat in the ocean watching a sunset” and I was rewarded with a result that was pretty much what I wanted to see. One of the details that I really like about this picture is the hair on their heads sticking up, something you would expect if you have ever been on a boat. Also, I like the perspective seeing them from behind, the ocean, the boom sticking out and lines, all of which are realistic.
Midjourney 5 Image Result After a Spelling Error
I decided to change course (yes, pun intended after a boat pic) to see how Midjourney 5 did when we see the faces and body of actual people. I prompted “a lady lounging by the pool,” and it generated the 4 images below.
Midjourney 5 Image Result After Correcting a Spelling Error
I will admit I was perplexed, as these were not at all what I requested. Then I re-read my prompt and saw I had typed “lunging” instead of “lounging” lol. So, it did exactly what I requested, which admittedly caused me to laugh out loud.
After that, I quickly corrected my error in a new prompt, and the result was as expected, though it feels too perfect. The water is too smooth. The lady is in an awkward pose and appears stiff. The plants, umbrella, everything just looks a bit fake like they were airbrushed or something.
Midjourney 5 Image of Basketball Players During March Madness
Then I then decided I wanted to see an action picture. Since it is March and March Madness is going on, I prompted “basketball players after hitting the game winning shot and celebrating on a basketball court.” The two images below were exactly what I wanted to see.
But for the first time, I think we are seeing some of the flaws still in the system. The legs are not proportional, blurry and awkward. If you look closely at the mouths, they don’t match the face correctly.
Still, it looks very realistic until you look at the details and way better than the previous version. But we can see there are still some additional steps forward that need to be taken for the AI to truly produce real action shots that are unrecognizable from the real thing.
Midjourney 5 Image of a Human Face
Next I wanted to see close ups to discover how accurately Midjourney 5 could create a human face. I prompted “an Irish woman with red hair and traditional dress,” and it generated the images below. Again, they are a bit too perfect. But who knows, maybe it applied a filter because the majority of photos on social media do these days lol. I do like how the hair is blowing in the breeze in the 2nd photo.
Midjourney 5 Image of a Human Face with Cultural Clothing
I decided to give the AI a challenge that was not only a human face, but also with traditional cultural clothing. I prompted “a close up of a Sioux tribe chief wearing traditional headdress and clothing.” I think these AI-generated images are my favorite of all of them. Midjourney’s AI did an amazing job with details, and the face looks so realistic. See what you think.
Midjourney 5 Image of Abstract Interpretation
Trying something different, I decided to test the AI with something more abstract, something it would have to interpret. I prompted “Mario from Donkey Kong in the real world,” and it generated the images below. I was very disappointed with the results. They are fake and video game-looking instead of translating Mario to the real world like an actual person.
I was not deterred though. I have worked with AI prompting long enough to know that when one angle doesn’t work, change to a different angle. Next, I prompted “a person in a Halloween costume dressed as Mario from Donkey Kong,” and it paid off with this much more realistic image.
Midjourney 5 Image of Translation of Video Game to the Real World
Finally, I decided to try again to have the AI translate a video game to the real world by taking another angle. I wrote the prompt “Joel from Last of Us on a movie set firing a gun at a zombie.” The results are not too shabby.
Conclusion
Overall, I am incredibly impressed with Midjourney 5. It’s a huge step forward towards AI-generated images that are so real you will not be able to tell if they are real photos or not. I encourage you to give Midjourney 5 a try. Let me know your thoughts, and I would love to see what you have Midjourney 5 create!
Contact the Author
Please feel free to contact me.
Luther Andal
[email protected]
Please contact us at your convenience to speak to someone at Streamline about a project.
