Popular AI Image Generators, “Midjourney” and “Stable Diffusion.” Who are giving attention?

AI has become an indispensable part of our lives, supporting us in a variety of areas, including Google's search engine and the video recommendation feature.

Especially in AI technologies related to images, image recognition has made great progress through deep learning technology. Simply put, image recognition is a technology for distinguishing an object based off of an image. AI image recognition has been applied to autonomous driving and facial recognition on smartphones.

Now, in 2022, AI Image Generators, which generates "pictures" from keywords, have been appearing one after the other. Its ability to anyone to allow anyone to "create a picture from keywords" has shocked many of its users.

For example, entering the keyword "The Gleaners, Millet" in the AI Image Generator "Stable Diffusion" will output the image below. The composition is completely different from the real one, but it depicts the scene of people gleaning stray stalks of wheat with a touch of oil painting style, just like the "The Gleaners" artwork.

Not everyone can create such a painting with a highly-quality finish. However, it is now possible to create works of art with the help of artificial intelligence. In this issue, we will investigate "Midjourney" and "Stable Diffusion," two of the most famous AI Image Generators.

“Midjourney" and "Stable Diffusion," released in June and August 2022, respectively, are AI that can generate images. Previously, there were already AI Image Generators such as "Imagen" developed by Google and "DALL-E" released by OpenAI.
Nevertheless, "Midjourney" and "Stable Diffusion," along with "DALL-E2," have attracted so much attention that they are called the "Gosanke," or "top three" AI Image Generators. This is due to their high performance, "capable of generating art that is comparable to that of those drawn by humans," and the “ease with which they can be used by anyone.”

One incident that is a great example exhibiting the high-quality performance of AI Image Generators is when the "Theatre D'opera Spatial," a painting created by “Midjourney” won an art contest. After winning the prize, the artist announced that it had been created by an AI Image Generator, drawing controversy.
In addition, "Artificial Images," a collection of artworks created exclusively by “Midjourney” and “Stable Diffusion,” was going to be sold starting September 2022. The momentum of "AI paintings" shows no sign of abating.

Then, what kind of users are searching for and using "Stable Diffusion" and "Midjourney"? We analyzed user attributes using Dockpit, VALUES' online behavior analytics tool, which can help analyze competitor websites and investigate trends based on behavioral data that is updated monthly.

First, how frequently are people searching for these AI Image Generator brands?
MANAMINA ranks the words that had the highest weekly spikes in searches and distributes them every Monday (except for public holidays). For July 31 - August 6, 2022, "Midjourney" ranked, and for August 21 - August 27, 2022, "Stable Diffusion" ranked.

Thus, by following the search terms that suddenly spike with popularity on a weekly basis, you can quickly and easily identify trends.

Looking at the trends in the number of searches, the number of searches for "Midjourney" and "Stable Diffusion" actually exploded in August 2022, with as many as 550,000 people searching "Midjourney." On the other hand, as for "Midjourney," the number of searchers decreased in September, suggesting that although it became a hot topic among information-sensitive users who take interest in new things, there are not many new searches since then, demonstrating the slowing down of the spread of its awareness.

Let's take a look at the keywords "Stable Diffusion" and "Midjourney" to see how they are perceived by users.

A high percentage of those searching for both "Stable Diffusion" and "Midjourney" are male. In particular, of those searching "Stable Diffusion," more than 70% were male, indicating that it is more popular among males.

Also, looking at the search trends by age group, we can see that the 20s and 30s age groups have the highest number of searches. It seems that "Stable Diffusion" is more popular with young people.

From this, we can say that both "Stable Diffusion" and "Midjourney" are searched mostly by "men in their 20s and 30s." What other characteristics do these users have? For further analysis, we used the behavioral analytics tool story bank.

The first characteristic to be mentioned is the length of time spent on Twitter.

The graph shows that more than 80% of those who searched for "Stable Diffusion" or "Midjourney" use Twitter, while less than half of all Internet users use Twitter. They often spend more than one hour on Twitter, which is significantly longer than the average of all Internet users.

This leads us to believe that in many cases, these AI Image Generators were first recognized on Twitter. It is also likely that users of each AI Image Generator are searching for keywords through the hashtag searches on Twitter to generate ideal images or sharing the images they have generated themselves. These "elements that make people want to share on social media" may be the reason for its popularity!

To gain a better understanding of users, we examined the interests of those who searched for "Stable Diffusion" and "Midjourney" in story bank in a similar fashion.

Both of those searching "Stable Diffusion" and "Midjourney" exhibited greater interest in "manga," "anime," and "games," possibly due to the popularity of AI Image Generators among young males.
On the other hand, those searching "Midjourney" were highly-interested in "art." The feature values for "art" are 28.60 points for “Midjourney” and 6.81 pts for "Stable Diffusion.”

We tried to find out why there is such a difference between the AI Image Generators by actually using both “Stable Diffusion” and “Midjourney” for ourselves.

There are two main ways to use “Stable Diffusion,” either by using the "demo website" or by using the "Oekaki Bari Gutto-kun" service on LINE.

Using the "demo website" is a recommended method for those who want to try using it once and do not want to go through a tedious implementation. On the other hand, there are some disadvantages, such as taking a few minutes to create and errors that occur during busy times.

For this method, you can use the service up to ten times a day for free by registering on “Oekaki Bari Gutto-kun” as a friend on LINE.

While this one is attractive because of the short time it takes to generate images and the ease with which it supports Japanese input, keep in mind that there are limitations for free use.

In addition, to use Midjourney, you will mainly use "Discord."
You can use Midjourney by logging in to Discord and pressing "Join the beta" on the Midjourney website. In this case, free use is limited to about 25 images per account, and if you want to generate more than that, you must register for a paid plan.

At this point, the preparation is complete. Now all you have to do is enter the keywords, the so-called prompts or “magic words,” for the art you want to create, and the AI will generate the art.

Here are some actual artworks generated by AI.

When comparing the two AI Image Generators, “Stable Diffusion” is more realistic and has an interesting mix of depictions that could not happen in reality, while Midjourney is not good at generating pictures of people or creatures, but is good at creating backgrounds and the atmosphere of the image.

The realistic nature of “Stable Diffusion” leaves little room for artistic thinking, which may be the reason for the low level of interest in "art" among its users.

Considering the possibility that the unrealistic expressions of "Stable Diffusion" and the abstract expressions of "Midjourney" could stimulate the imagination of painters and provide inspiration for their hand-drawn paintings, we felt that it is unlikely that the technology will take away the jobs of artists. On the other hand, it must be understood that for the painters/artists whose paintings were used for training of the AI, they are bound of have negative feelings about the AI generating images similar to their artworks.

One of the problems with AI Image Generators is that it is not good at generating a group of images with a storyline. The principle of AI Image Generators is to repeatedly determine whether the trial image resembles the input keyword (prompt) and improve the accuracy of the image. This means that even if the prompts are the same, the output images will not be the same. In other words, it is impossible to generate "a set of four images of the same boy with expressions of joy, anger, sorrow, and pleasure" as shown below. Thus, it is difficult to give a "narrative" to the images, and it is probably one of the shortcomings of AI that it can only generate and depict one fragmentary situation.

“Stable Diffusion” and “Midjourney” were investigated to better understand AI Image Generators.

We can take the following three implications from the study:

・Men in their 20s and 30s were the most likely to search for AI Image Generators, and their time spent on Twitter is relatively long.
・“Stable Diffusion” excelled at realistic expressions, while “Midjourney” was not good at generating portraits and was more abstract.
・The AI Image Generators had difficulty in creating a group of images with a storyline.

In the future, surely a market for AI Image Generators will further develop. One of the keys to expanding awareness of AI Image Generators will be to create a model that does not exclusively target young people, and to consider how to get those less interested in AI to accept AI-generated artwork, which will be essential to market expansion.

▼Data from Dockpit was used in this case study. Dockpit is an online behavior analytics tool provided by VALUES, Inc., and it allows you to access online behavioral data (updated monthly) via your browser so you can research competitor websites and investigate trends. There is a FREE version of Dockpit, so if you are interested in using it, please register below.