My MidJourney experience: Creating AI-generated pictures from text input

Just awesome

I can't find other words than this to summarize the entire experience so far: my brains have been blown away! The amount of detail MJ is able to put in each image is astounding.

The process is as follows. You can join their Discord server, and then by using a simple prompt send in a command with some text and options for image generation, and within 20-30 seconds, you get four images from MJ to choose from which you can enlarge.

Take for example the image below that has been generated by simply using the following command: /imagine beautiful blue universe.

Look at how amazing the image looks like. Although some details might not always match reality, it comes pretty close to ... well, some kind of reality.

Other applications

Except for the nice imagery, what I also found interesting is to explore different logo's for my website. By using specific text and telling MJ that you want a logo generated, you get pretty interesting and detailed designs. Let's try.

/imagine professional brand logo in square box, black and white, subtle, simple.

For logo creation this enables you to test out different ideas quicly. MJ can really help you brainstorm on different visual concepts and ideas. You can combine different colors, forms, and also create many variations until you find what you like.

But it does not stop here. Let's say you're working on your CV and want to give a refresh on the design. You can ask MJ to give you some ideas. Check this out:

/imagine detailed cv template, black matrix green

Mix-in ideas into new dimensions of creativity

There is so much more to explore here. Once you've found something you like, you can also mix different ideas together and let MJ do the work.

Let's say you're a concept artist and want to have a concept drawing of an alien ship. You can do something like this:

/imagine alien ship in the style of Mark Ryden, b&w illustration on paper

For more realistic concept designs of 3D art, you can also try something like this:

/imagine futuristic sleek space ship hyperspace, ultra detailed, hyperrealistic

What about super futuristic UI interfaces? No problem.

Maybe some new superhero for your Unreal Engine game you're working on. You want a hyperrealistic model and want some ideas on how such a dude would look like. To spice things up, there is blue fire coming out of his head. Got ya.

Or maybe you're working on a space-sim game or sci-fi novel where the setting is a highly advanced military base on some golden planet far away. Yep, can do this as well.

Not perfect (yet), but great for prototyping, brainstorming and experimentation

Except for the amazing images that MidJourney creates, there are so many applications to these works as well. You can brainstorm different ideas and stimulate your creativity in many unexpected ways.

However, MidJourney does not do a perfect job (yet). I found that especially faces are difficult to render and it very often misses out on specific details. In other situations some details just doesn't make sense where they don't reflect reality. Take for example the below picture of "rabbit hole portals" try-out with something in the middle that should be a rabbit.

But bare in mind that as of this writing, MidJourney is still in beta. I expect that over the next iterations of the machine learning model, this will improve drastically.

Conclusion

Overall, I'm totally impressed. I honestly didn't expect this level of detail and creativity coming out of a machine learning model. This is actually so good already, that I expect a total disruption of designer/creative industry in the coming years. There are already many designers that tried out models like MJ and DallE and started worrying how this would impact their own work.

Mid Journey doesn't just generate images. It does not just generate art. It generates inspiration and creativity. It stimulates your imagination to get to places where you have never been. This is some next-level shit.

Let's end this post with some more images: