MidJourney is an AI tool that utilizes natural language processing to generate visually stunning images from text inputs. Unlike Dall-E, which is primarily focused on producing realistic images, MidJourney is geared towards creating artwork with an emphasis on aesthetic appeal. The impressive results have captivated many artists and enthusiasts alike, making it a popular topic of discussion. In contrast, Dall-E Mini alias Craiyon gained viral success on social media since June 2022, as it can generate funny memes and images based on user-generated texts. However, a few weeks later, MidJourney emerged as a new text-to-image generator that is generating a lot of buzz on the internet.
Contents
- What is MidJourney?
- What is MidJourney used for?
- How to create images with MidJourney?
- What content is prohibited?
- Who owns the intellectual property of the art created by MidJourney?
- David Holz: who is the creator of MidJourney?
- Is Midjourney free?
- MidJourney vs. DALL-E
- The phenomenon of AI-generated art
- The future of MidJourney
- MidJourney’s best creations
- MidJourney V4: all the new features
- MidJourney bans words related to the reproductive system to avoid porn content
- Cyberpunk: Peach John, the first manga created with MidJourney
- MidJourney goes to version 5: photorealism and (almost) normal human hands
- MidJourney Launches Entirely AI-Created Magazine
- MidJourney becomes paying, the fault of cheaters
- MidJourney bans images of Xi Jinping and creates controversy
- The new “/Describe” feature converts an image into a prompt
What is MidJourney?
MidJourney is an AI system that enables users to generate images from textual inputs. Its creators describe themselves as an independent research laboratory that explores new avenues of thought and extends the limits of human imagination. Unlike other text-to-image models, MidJourney was trained on an extensive dataset of images, but its primary focus is on artistic style rather than realism. The creators of the AI optimized it to generate visually pleasing images.
Currently, MidJourney is in version 3.0 and features a feedback loop based on user activity and reactions. By analyzing data on user preferences and image usage patterns, the software can enhance image quality, resulting in more impressive and aesthetically pleasing outputs.
While the specific technologies used by MidJourney have not been disclosed, the company has confirmed that it utilizes the latest and most advanced AI models with billions of parameters, trained on a vast dataset of billions of images. The images are generated on a cloud vendor’s platform that utilizes green energy.
Generating each image using MidJourney requires several petaops, which translates to 10^15 operations per second. According to the creators, no publicly accessible service has ever provided an individual with access to such a significant amount of computing power.
What is MidJourney used for?
While many users employ MidJourney for recreational purposes and to bring their ideas to life, approximately 30% of users utilize the tool for professional purposes. Graphic designers, for example, use MidJourney to create various design concepts and present them to their clients, who then select the direction they want to pursue. According to the creator, MidJourney “supercharges” the creative or communication process for professionals.
In addition, about 20% of users use MidJourney for therapeutic purposes, as creating images can help them overcome trauma or grief. In this way, AI serves as a tool for emotional and intellectual reflection.
Previously, MidJourney was available only for non-commercial use. However, commercial use is now authorized. In fact, in June 2022, the prominent publication The Economist used AI to create its cover image.
How to create images with MidJourney?
MidJourney is still in beta and interested users need to register on the official website. Invitations are sent out in batches to registered users.
After receiving an invitation, users can communicate with MidJourney via a public channel on Discord. They simply need to enter some text and the MidJourney robot will return an image after approximately a minute.
Users can also opt for a paid subscription of either $10 or $30 per month, which allows them to submit their text in private through a direct message to the robot. This eliminates the need to share messages on the public channel with other users. However, it’s worth noting that the images generated by the AI are still publicly visible by default.
What content is prohibited?
MidJourney has strict guidelines and limitations on what kind of content is allowed on the platform. Gore or adult content is strictly prohibited, and users are urged to avoid creating visually offensive or disturbing content.
Certain texts that violate the platform’s guidelines can be blocked automatically, and the platform has around forty moderators who monitor the images created by users to ensure compliance.
The creators of MidJourney are also opposed to their AI being used to create fake DeepFake-like photographs. They believe that this is an “extremely dangerous” use of the technology.
Who owns the intellectual property of the art created by MidJourney?
The issue of intellectual property when it comes to art generated by AI is a sensitive topic. Currently, in the United States, copyright law prohibits granting copyrights to images generated by AI.
In February 2022, the U.S. Copyright Office Review Board denied a motion to copyright a computer-generated landscape image titled “A Recent Entrance to Paradise”. This decision was based on the fact that the image had not been created by a human, and therefore did not meet the criteria for copyright protection under current law.
Intellectual property issues are indeed a complex topic in the context of AI-generated content. MidJourney’s approach to requiring a copyright license from users to reproduce content is a common practice among companies that host user-generated content.
Regarding AI models trained on copyrighted material, there is a risk of infringing on the original copyright. However, the legal framework for this issue is not yet fully developed, and there is ongoing debate about how to address it.
Some argue that AI-generated content should be treated similarly to traditional copyright, while others suggest that new forms of legal protection may be necessary. It is likely that laws and regulations will continue to evolve as the use of AI in creative fields becomes more prevalent.
David Holz: who is the creator of MidJourney?
David Holz, a Florida native, started his design business while pursuing math and physics degrees at university. Although he was on track to complete his PhD in Mathematics, he took a break in 2008 to co-found Leap Motion, a company that manufactured computer peripherals. Holz spent time as a research student at the Max Planck Institute and as a student researcher at NASA’s Langley Research Center, working on various projects, including LiDAR, Mars missions, and atmospheric science. However, Holz began to question why he was working on these projects and decided to focus on his passion for Leap Motion, which develops a hardware device that tracks hand movements and uses them to control devices.
After running Leap Motion for twelve years and employing about 100 people, Holz left to create MidJourney in 2021. The company currently has around ten employees and is self-funded with no investors. Holz notes that the team’s main motivation is not financial gain but rather working on projects they are passionate about and having fun. MidJourney works on various projects related to AI technology, but Holz emphasizes the difficulty in predicting the human ramifications of its expansion.
He believes that the intersection of humanity and technology is an important area that requires extensive experimentation to understand fully. Although the company is already profitable, Holz believes that there is much to learn about the impact of AI on society and how best to use it for the benefit of all.
Is Midjourney free?
Midjourney provides a free trial service where users can create 25 images for free before choosing between two subscription packages. The basic subscription costs $10 per month and allows users to create up to 200 images per month. Once users exceed the 200 image limit, they can generate additional images at a cost of $4 per 60 images. The basic subscription also includes free friend passes and member gallery access.
The standard subscription offers unlimited personal users, free passes for friends, and access to the members’ gallery. Additionally, it includes 900 minutes of accelerated time to help users generate images quickly. Users can add a private view option to both subscription packages for an extra $20 per month. Plan changes can be made at any time, but Enterprise plans cannot be downgraded.
MidJourney vs. DALL-E
While DALL-E prioritizes the realism of generated images, MidJourney focuses more on the artistic dimension by adopting real art styles to create images from desired elements. In a comparison test conducted by user Fabian Stelzer, the same text was inputted into both programs. Stelzer found that MidJourney had a certain “je ne sais quoi” with imperfections that are more beautiful, similar to analog synthesizers. MidJourney was also noted for its contextual creativity and excellent use of textures. DALL-E was found to excel in scenes with clear instructions.
Test texts included “Mozart playing Top of the Pops, 1993” and “a complex installation made of plastic bags and mirror fragments, painted in neon color, studio lighting.” In the case of the installation made of plastic bags, MidJourney was able to create a more realistic image with successful lighting of the studio.
The phenomenon of AI-generated art
The emergence of AI-generated images can be traced back to 2021 when OpenAI launched CLIP (Contrastive Language Image Pre-Training), which was initially designed to evaluate how well generated images matched textual descriptions. However, artist Ryan Murdock realized that the process could be reversed, and an AI could create an image from text. This sparked a wave of exploration in the generative art community, with Python code being released for creating images using various patterns and techniques.
David Holz, the creator of Midjourney, noted that around this time, “some areas of AI were progressing in really interesting ways, and one of them was the ability of AI to understand language.” He specifically referred to transformers, a type of deep learning model on which CLIP is based. Additionally, diffusion models are an alternative to GANs (generative adversarial networks).
Holz was particularly impressed with the diffusion guided by CLIP, which was developed by Katherine Crawson, known on Twitter as @RiversHaveWings.
The future of MidJourney
As the technology behind MidJourney continues to evolve, its creators hope to continue pushing the boundaries of what AI can do in the field of generative art. They believe that there is still much to be discovered and developed in this area, and are excited to see what the future holds.
One area of focus for the team is improving the quality of the generated images, with a particular emphasis on realism and detail. They also hope to make the technology more accessible to a wider range of users, including those who may not have a background in art or programming.
At the same time, the team is aware of the potential risks and challenges that come with the increasing use of AI in creative fields. They are committed to being responsible stewards of this technology, and to ensuring that it is used ethically and with appropriate safeguards in place.
Ultimately, the goal of MidJourney is to empower artists and creatives to explore new frontiers in their work, and to unlock new possibilities in the world of generative art. Whether it succeeds in this mission remains to be seen, but there is no doubt that the technology behind MidJourney is already pushing the boundaries of what is possible with AI and machine learning.
MidJourney’s best creations
By creating a MidJourney account, you can view other users’ creations on the Community Feed page. New AI-generated images are constantly pouring in. Here is a selection of impressive or surprising creations made by MidJourney and relayed on social networks…
MidJourney V4: all the new features
MidJourney started the alpha test of its V4 on November 5, 2022, which promises more comprehensive results and streamlined access to high-quality output from simple texts. This latest iteration replaces the previous V3 released in August and is being tested by thousands of members of the official Discord server who can access it by simply adding “-v 4” to their prompts. The V4 represents a significant upgrade as it features a completely new codebase and AI architecture trained on the new AI Midjourney supercluster developed over a nine-month period, according to founder David Holz.
The V4 model produces images that are significantly more detailed, with improved scene composition and realistic proportions. It also demonstrates a greater knowledge of places, creators, and other details, with better representation of small details and support for complex promptings with multiple levels of detail. The model is also better suited for scenes with multiple objects or people and offers advanced features such as multi-prompt and image prompting. Users have warmly welcomed this update, which is currently in the alpha stage and will continue to be refined in the coming weeks. Future improvements planned by the company include increasing the definition and quality of upscaled images, adding custom aspect ratios as in the V3, improving image precision, and reducing text artifacts.
MidJourney bans words related to the reproductive system to avoid porn content
MidJourney has always had strict policies prohibiting the generation of pornographic content on its platform. However, some users have found ways to circumvent these restrictions. To combat this, the company has decided to ban words related to the human reproductive system, such as “placenta,” “fallopian tubes,” “mammary glands,” “sperm,” “uterine,” “cervix,” “hymen,” and “vulva,” as prompts for generating content. Users attempting to use these words may be temporarily blocked from the platform. According to David Holz, this is a temporary measure to prevent the creation of violent or sexualized images while the company improves its AI. The list of banned words can be found on MidJourney’s terms of use page.
Cyberpunk: Peach John, the first manga created with MidJourney
The upcoming release of the manga “Cyberpunk: Peach John” in Japan has caused a stir, as its author, Rootport, admits to having no talent for drawing. Instead, he turned to generative AI tools like MidJourney, Stable Diffusion, and DALL-E 2 to create a manga of over 100 pages in just six weeks. With a prompt containing keywords like “pink hair,” “Asian boy,” or “stadium jacket,” Rootport was able to quickly generate a hero for his story, and he then assembled the best images into a comic book format. The book will be published by the renowned publisher Shinchosha on March 9, 2023.
According to Rootport, generative AI could allow many people without artistic talent to enter the manga industry, as long as they have good stories to tell. He likens the creative process to a lottery or the incantation of magic spells, allowing him to generate images from his imagination. However, he admits that the satisfaction of drawing oneself is undoubtedly greater.
This first AI-drawn manga raises concerns for the multi-billion dollar manga industry, as some fear that the junior assistants of mangakas may be replaced by AI. Professor Satoshi Kurihara expressed this concern to AFP.
MidJourney goes to version 5: photorealism and (almost) normal human hands
MidJourney made an announcement on Wednesday, March 15, 2023, about the release of version 5 of its text-to-image service. The new version is capable of producing photorealistic images, to the extent that some users find them almost “too perfect”.
The new version 5 of MidJourney’s text-to-image service shows significant improvement over the previous versions launched in March 2022, August, and November of the same year. This latest iteration generates highly realistic images, with better skin textures and facial features, and more accurate lighting, highlights, shadows, and viewing angles. The level of detail has increased significantly compared to previous versions.
In addition to the significant improvement in skin textures, facial features, lighting, and viewing angles, the latest version of MidJourney’s text-to-image service also shows progress in other areas. For example, the eyes are now much more realistic, and the hands are much closer to being normal, which was one of the main weak points of the tool in the past. Instead of having 7 or 10 fingers, as was often the case before, hands now have 5 fingers most of the time, which is a significant improvement.
In addition, MidJourney specifies that v5 offers a much wider range of styles and is more sensitive to prompting. It also generates less unwanted text and offers twice the image definition .
MidJourney v5 is available in alpha testing for subscribers to the service. It can be accessed from Discord.
MidJourney Launches Entirely AI-Created Magazine
After gathering input from its users, MidJourney has decided to release a monthly print magazine with the same name as the platform. Every issue will showcase a collection of images produced using AI and highly praised by the community. Additionally, the publication will feature sample prompts and interviews with the creators of the images. Interested readers can subscribe for $4 per month, while early adopters will receive the debut issue at no cost.
MidJourney becomes paying, the fault of cheaters
On March 30, 2023, MidJourney decided to end the free trials of its service. CEO David Holz cites “ the extraordinary demand and abuse of testing ” as the reason.
According to him, this break is due to “ a massive number of people creating disposable accounts to get free images ”.
He also specifies that the culprit could be “ a tutorial video that has gone viral in China ” . However, this wave of abuse occurred at the same time as a “ temporary shortage of GPUs ” .
The combination of these two hazards caused a service outage for paying users . This explains why the firm preferred to simply remove the free trials.
The company says it is still looking for a way to offer free trials again . In particular, it tried to require an active email address, but that was not enough and it will therefore be necessary to find a viable alternative…
MidJourney bans images of Xi Jinping and creates controversy
MidJourney appears to have a policy of restricting the use of certain prompts containing the name of Xi Jinping, which some users view as a form of censorship that contradicts the platform’s commitment to freedom of expression. While the platform allows DeepFakes of other prominent figures like the Pope, Macron, and Donald Trump, any attempt to generate an image using a Xi Jinping prompt triggers a warning message. In June 2022, MidJourney CEO David Holz explained that this was done to prevent scandals and dramas, and that political satire is received differently in the West and China, where it can be dangerous for users.
The new “/Describe” feature converts an image into a prompt
The new image-to-text feature launched by MidJourney in early April 2023, known as the “/describe” command, offers several advantages for various use cases. One of these is improved web accessibility, particularly for the visually impaired who can now better understand what is displayed on their screens. Before this tool, visually impaired individuals had to rely on manually created “alt text” descriptions associated with images.
MidJourney’s automation of image description tasks not only saves valuable time but also enhances accessibility and search engine efficiency through AI-generated comprehensive descriptions. This, in turn, improves the web as a whole by providing more context and explanation through captions on images. Additionally, the image-to-text generation creates a feedback loop for MidJourney’s text-to-image system, allowing users to create variations of existing images using the AI-generated descriptions as prompts.
With MidJourney, users can access four different descriptions for each uploaded photo, and they have the option to generate new variants or manually edit them. To upload an image, simply type “/describe” in the text field, and a drag-and-drop upload field will appear.
