Midjourney is a generative AI program and service created by Midjourney, Inc., an independent research lab based in San Francisco, California. The platform generates images from natural language descriptions, called prompts, in a manner similar to other Artificial intelligence image generators. It became one of the prominent technologies of the recent AI boom. The tool entered open beta on July 12, 2022, and is accessible through a Discord bot or an official website, where users can produce images without needing specialized hardware or machine-learning expertise.
The project is led by David Holz, a co-founder of Leap Motion, a company known for hand-tracking sensors. In an August 2022 interview with The Register, Holz stated that Midjourney was already profitable, a distinction among many early AI ventures. Users interact with the service by sending commands through Discord or using the web interface, with the platform returning sets of four images for each prompt, which can then be upscaled or refined.
History and development
Midjourney, Inc. was founded in San Francisco by David Holz, who previously co-founded Leap Motion. The Midjourney Discord server launched on March 14, 2022, with the team requesting high-quality photographs to be posted to Twitter and Reddit for training purposes. The company began releasing new model versions every few months, starting with version 2 in April 2022 and version 3 on July 25, 2022. The alpha iteration of version 4 was released on November 5, 2022, and from that version onward, models were trained on Google TPUs instead of standard GPU clusters.
Version 5 entered alpha on March 15, 2023, followed by the 5.1 model, which applied more stylization to images and worked with more literal prompts. The 5.2 version introduced an aesthetics system and a "zoom out" feature that generated surrounding content for existing images. Version 6 reached alpha on December 21, 2023, trained from scratch over nine months, with improvements in text rendition and a more literal interpretation of prompts. The web interface was launched in August 2024 alongside version 6.1, consolidating editing, panning, zooming, region variation, and inpainting into a single tool.
Features and functionality
Midjourney is primarily accessed through a Discord bot, either in the official server, directly messaging the bot, or by inviting it to a third-party server. Users type the "/imagine" command followed by a prompt, and the bot returns a grid of four images with options to upscale. The website initially required users to generate at least 1,000 images via Discord before using the web editor, but that limitation was removed. The platform offers multiple refinement tools, including Vary (Region) and remix, which was released on September 5, 2023, allowing users to select a specific area of an image and apply changes only to that region.
The Image Weight feature controls the influence of an uploaded image on the final output, letting users prioritize either the prompt text or the image characteristics. The Style Reference tool extracts colors, textures, or atmosphere from a reference image and applies them to new creations. The Character Reference feature uses an uploaded character image to maintain visual consistency across multiple generated images of that character.
Use cases and adoption
David Holz has described how artists use Midjourney for rapid prototyping of artistic concepts to show to clients before beginning a final render. The advertising industry adopted the tool, along with DALL-E and Stable Diffusion, for brainstorming ideas and creating original content. Architects have reported using the software to generate mood boards for the early stages of projects as an alternative to browsing image search results. The platform has also been used in media production, including magazine covers and comics.
Notable controversies
The BBC noted in 2023 that the realism of AI-based text-to-image generators, including Midjourney, had reached a high level, prompting cultural debate. One prominent case occurred in August 2022 when a Midjourney image titled "Théâtre D'opéra Spatial" won the digital art competition at the Colorado State Fair. Jason Allen, who wrote the prompt, printed the image on a canvas and entered it under his name, a move that angered other digital artists. Allen was unapologetic, stating that he followed the competition's rules, and the judges later said they would have awarded the prize even if they knew about the AI tool.
In December 2022, the AI-generated children's book "Alice and Sparkle" was created over a single weekend. Its creator, Ammaar Reeshi, used Midjourney to generate around 13 images for the book, which drew criticism from artists who alleged that the model was trained on their works without permission. Charlie Warzel, writer for The Atlantic, used Midjourney to generate images of Alex Jones for a newsletter, later calling the decision a mistake in an article about the incident. The program was also used by The Economist for a 2022 front cover and by the Italian newspaper Corriere della Sera for a comic written by Vanni Santoni. A 10-minute segment on Midjourney was broadcast in an episode of Last Week Tonight with John Olliver.
Reception and criticism
Artists have raised concerns about consent and credit, as Midjourney's training data includes many copyrighted works. While some praise its accessibility and speed, others criticize the legal and ethical implications. The company has responded by adding features like the option to opt out of training data, though such measures have been viewed as insufficient by critics. Despite the debate, Midjourney remains widely used professionals.