Midjourney (www.midjourney.com) is a neural network that generates art from English phrases or phrases in many other languages. It is an exciting tool to convert your thoughts into art. The limit is only one’s imagination.
How to use Midjourney
To use midjourney, one needs to have a discord account, which is free. One can create it easily with a few details such as name, email and date of birth.
Then one needs to go to the discord server for midjourney (https://discord.gg/midjourney) and join it.
Common midjourney commands
One should start with any of the newbie rooms in the midjourney discord.
The commands are
/imagine<phrase>
This command generates an image corresponding to the phrase using midjourney AI
/info
This command gives information about the user’s subscription (free or paid) including how many images are remaining till the subscription runs out.
/subscribe
This takes one to the subscription page where one can opt for any of the paid subscriptions.
There are various options for size, resolution, art style (such as picasso style or comics style or ghibli to van gogh style or moebius style) that one can provide along with the /imagine command.
One can read the quick start documentation at https://midjourney.gitbook.io/docs/
Free and paid subscription
One can get around 25 to 30 images generated with the free account.
After that one needs to go for one of the paid services, starting from the basic subscription at $10 per month for 200 images.
Some midjourney generated images
Below are some of the images I generated
Some of the imagine commands (you could try these after the /imagine) are as follows:
- Mars colonized with forests and people, moebius style
- Buddha teaches disciples, van Gogh style
- End of the world, comics style
- Battle of Panipat, anime style
- turkey grand bazaar, hyperrealistic
- The battlefield in Battle of Talkatora, War between Vijayanagar and Bahmani kingdoms, photorealistic
Things to note about midjourney
The midjourney software is usually better at generating images of landscapes and first person. It is perhaps less good at capturing specific things such as particular events or interactions between people.
Midjourney is not intelligent and does not get irony: so one has to be very precise in describing the desired scene or image in as simple words as possible.
One can choose to get variations of the original image, or upscale the resolution to make it much more detailed.
Note: you have to be a little careful about the words you use to generate the images, or you could get banned. Also, sometimes the AI isnt ‘clever’ enough to understand what you mean exactly, so again you have to choose the words more carefully.
Examples of Midjourney fails
Below are some examples of midjourney fails where the AI completely didnt understand what I wanted it to do, or I might have not chosen the correct words or phrases.
Example 1: Mehmet conquers Constantinople: This is a fail because before it was conquered, Constantinople had big walls and did not have the minarets. So it could not make the historical connection.
Example 2: Taj Mahal, melancholy, Van gogh: Thats not how Taj Mahal looks, although the Van Gogh style is good.
Example 3: Hannibal Lecter eating human body parts: Didnt quite get it
Example 4: Coding interview with Google: not quite!
Again, it just shows one needs to be more skilled at choosing the words. As for the factual errors, hopefully midjourney will become better as the training improves.
Conclusion
In this article we gave a brief introduction to midjourney and how to use it to generate images from phrases.
