Midjourney: The gothic AI image generator challenging the art industry – BBC Science Focus Magazine
Yet another AI image generator that is taking over the internet, Midjourney is all about using your imagination.
Artificial intelligence (AI) has been given a new task in recent times – to take over the art world. Following the huge success of OpenAI’s Dall-E 2, there has been a sudden rush of AI programs looking to turn the layman into a modern-day Picasso.
While there are plenty of companies to choose from, one of the biggest (and arguably most unique) generators is known as Midjourney. Unlike its pristine competitors like Dall-E or Google’s Imagen, Midjourney wants to add some flare to your prompts.
Dripped in dark tones, fantasy themes and an atmosphere Edgar Allen Poe would be proud of, Midjourney is the AI generator for creatives looking to guide their next book, game or science-fiction journey. But what is Midjourney, how does it work and can you use it?
Midjourney is one of the many AI image generators that have emerged recently. Unlike Dall-E 2 or some of its other competitors, Midjourney offers a more dream-like arty style to your requests.
It will likely appeal to those working within science-fiction literature or artwork that requires a more gothic feel. Where other AI generators lean more towards photos, Midjourney is more of a painting tool.
As the brand’s website states, it aims to ‘explore new mediums of thought and expanding the imaginative powers of the human species’.
Midjourney asks you to input a worded prompt for an image, for example ‘a fox wearing a top hat in the style of a Roald Dahl illustration’ and in a few seconds, you’ll be returned multiple attempts at this image.
Things can go wrong with your image and the model isn’t perfect. For example, it can struggle to generate hands or other complicated objects, and backgrounds can often be blurred or confused, but these are problems you’ll find with any of these models.
Midjourney works slightly differently from most AI tools. While you can sign up directly through the company’s website, using the software will require you to download and use Discord – a chat forum-style app.
Once you’ve been accepted onto Midjourney’s Beta, you’ll be sent an invite to the Midjourney Discord. This can seem pretty confusing if you haven’t used anything similar before but the basics are easy.
You get 25 free queries and then, if you want to keep using the software, you need to sign up for a paid membership (costing between $8 and $60 a month depending on the plan you choose).
When you’re in the Discord, simply head to any of the ‘newbie’ channels listed on the left-hand side. Then, type ‘/imagine’ followed by your desired prompt. The bot will respond with four versions of this prompt within a minute.
You can choose to get alternative versions of any of these images, or to upscale one of them to a larger, higher quality image.
In these bot channels where you request your images, other people will also be making requests. It can get a bit hectic in there so keep track of where your image is.
You can see any image you have requested on the Midjourney website.
For most of the major AI image generators, the creators have offered reams of information on how they work, along with publishing their source code – this isn’t the case for Midjourney. The team behind it have kept somewhat quiet on its background and training.
However, it is most likely that Midjourney uses a similar system to both Dall-E 2 and Stable Diffusion, especially as both companies have explained their training methods in detail.
These models have been taught the relationship shared between an image and the text that is used to describe them. Midjourney has stated in the past that, like its competitors, it scraped the internet for images and text to describe them, using millions of published images for training.
The majority of AI image generators with public source code generate these images through a process known as diffusion. Essentially, these models work by adding noise to an image, reducing it to a pixelated mess.
It then learns to recover this data by reversing the noising process. This process is repeated over and over again, training the model to add noise and then remove it again. The model can then apply the denoising process to create realistic images by making small variations to the image.
This is paired with the model’s understanding of the relationship between an image and the text used to describe it, helping AI image generators to understand what you are asking for with your prompt.
Like its competitors, Midjourney is not without criticism. There are concerns over copyright, especially since the founder admitted they didn’t receive consent from the creators of the art the model was trained on.
With millions of images needed for a model of this size, this would mean copyright infringements in the millions.
This has resulted in both Midjourney and Stable Diffusion getting caught up in a lawsuit brought forward by a team of artists.
"It’s [the dataset] just a big scrape of the Internet. We use the open data sets that are published and train across those. I’d say that’s something that 100% of people do. We weren’t picky," says David Holz, founder of Midjourney, in an interview with Forbes.
In the future, Midjourney are hoping to be able to offer artists the ability to opt out of their images being used in the future.
"The challenge now is finding out what the rules are, and how to figure out if a person is really the artist of a particular work or just putting their name on it," says Holz in his interview with Forbes.
"We haven’t encountered anyone who wants their name taken out of the data set that we could actually find in the data set."
Alex is a staff writer at BBC Science Focus. He has worked for a number of brands covering technology and science with an interest in consumer tech, robotics, AI and future technology.
Subscribe to BBC Science Focus Magazine and try 3 issues for just $9.95
Discover our latest special editions covering a range of fascinating topics from the latest scientific discoveries to the big ideas explained.
Listen to some of the brightest names in science and technology talk about the ideas and breakthroughs shaping our world.
Our daily newsletter arrives just in time for lunch, offering up the day's biggest science news, our latest features, amazing Q&As and insightful interviews. PLUS a free mini-magazine for you to download and keep.