Midjourney

Revision as of 15:30, 9 April 2023 by Daikon Radish (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Midjourney Photographers Styles Reference Sheet

Midjourney Artists Styles Reference Sheet

Overview

Midjourney is an artificial intelligence program capable of generating artistic images based on input text (i.e. prompts) [1] [2]. It is also the name of the overall project; according to their official website, “Midjourney is an independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species” [3]. The AI-generating software has been compared to OpenAI’s DALL-E and likened, in its present form, to a social network for creating art based on text descriptions [1] [4].

There are two ways to experience the program: one is the Midjourney Bot, which can be used on Midjourney’s official Discord server or any other Discord server where it has been setup, and the other is the web app available at www.midjourney.com. With the bot, a user is able to generate images and the web app allows users to see galleries with their own works [1] [3].

The AI art generating program is capable of associating words with specific image features, allowing users to write an artist’s name on the prompt to create an image in a style associated with that name or to insert words that describe types of artwork [5].

In 2021, after the release of OpenAI’s CLIP (Contrastive Language-Image Pre-training), designed to assess the alignment between text descriptions and generated images, there was an increase in popularity for creating high-quality images from AI models based on text input. This led to a period of exploration and development by the art community, sharing Python code and using various models and techniques to create images [1].

Midjourney’s technology is different from other image creation tools like Adobe Photoshop or Blender. It exists as a social app, using as its front-end Discord. Users log in to Midjourney’s server, submit text prompts, and an image is generated and shared within the server after, more or less, a minute of computation. Due to its social nature and its creator’s goal of it being a social space for anyone over the age of 13, it is subjected to rules about allowable content, with some text inputs being blocked automatically. The company has around 40 moderators monitoring on the user’s images generated content. For example, the creation of images with adult content or gore isn’t permitted. Also, it is advised that user’s refrain from making overall disturbing or shocking content [1].

The program output, like DALL-E, results from AI models training on a large number of images. The third version of Midjourney has been introduced with enhancements in image quality coming from it’s social aspect, incorporating “a feedback loop based on user activity and response.” According to David Holz, founder of Midjourney, “It's mind-bogglingly better and we didn't actually put any more art into it. We just took the data about what images the users liked, and how they were using it. And that actually made it better" [1].

Current Applications and Future

Midjourney has been adopted by both professional artists and amateurs. Indeed, the majority of people are using it for fun, engaging with their imagination. Some use it as a kind of art therapy, creating images of pets that have died, for example. Regarding professionals, architects and graphic artists use the program to help with the development of their concepts. It’s an easy and quick way to generate several variations of an idea to see what fits bests to the clients’ needs [1] [3].

It has also been used to create music videos. Shane 54, a Hungarian DJ, songwriter and animator spent months exploring AI image generating programs. As his interest grew, he wanted to apply this technology to his music by generating “every single element of the video with Midjourney, creating a collage that is coherent like no other. Then animate it” [6] [7].

During the 2022 Colorado State Fair’s fine art competition, an artist called Jason Allen won first place in the digital category with a work created through Midjourney entitled “Théatre D’opéra Spatial”. The image was printed on canvas for submission, depicting a scene where classical figures in a hall are looking through a circular viewport [8].

Indeed, the use of AI-generated artwork has caused concern among artists due to the possibility of it reducing job opportunities. Matt Borrs, a cartoonist, said about AI art that “To developers and technically minded people, it’s this cool thing, but to illustrators it’s very upsetting because it feels like you’ve eliminated the need to hire the illustrator.” Some magazines, like Atlantic and The Economist, have already used artwork generated by AI for their covers [8].

Although there has been adoption, the mass use of technologies like Midjourney, in their current state, still have the challenge of scaling. Presently, the service has hundreds of thousands of users, requiring around 10,000 servers. Increasing the numbers of user to the tens of millions would mean that there would be a need for more servers than there are in existence [1].

Development

Midjourney was founded by David Holz after leaving Leap Motion, a company he ran until 2021. In 2022, an invitation only private beta was launched around early March [1] [9] [10]. The open beta was announced on July 13th, 2022 [11].

Creating images with Midjourney

After joining Midjourney’s Discord server, the user will be able to generate art based on text prompts. At first, the program will present four diffused images as previews. After the user chooses the ones he wants, a higher resolution version will be created. There is a limit of 25 free uses of text prompts, using the /imagine command and other queries, without subscription [3] [10].

For a user to create an image, the official documentation recommends the following steps:

  • Select one of the #newbies channels from the sidebar;
  • Use the /imagine Command;
  • Wait as the Midjourney Bot Processes Your Request;
  • Upscale Your Image or Create Variations;
  • Save Your Image;
  • Send Results to Your DMs (Direct Messages) (You can ask the Midjourney Bot to send you a Discord direct message containing your final results.);
  • Become a Paid Subscriber if you Want to Generate More (Your 25 free jobs do not presently expire but will also not renew. Once you have used all of them, you can use /subscribe in any Bot channel to purchase a membership.)
  • As a Subscriber, Direct Message the Bot to work one-on-one with the Bot. (If you are a paid subscriber, you can use the bot in your Direct Messages instead of public channel with other people. Images made within your direct messages are still be subject to content and moderation rules and will be visible on your Midjourney website gallery.); [3]

There is also advice given regarding the text prompts. The user can be specific or vague while writing the text; a vague description will result in more variety but may miss the mark of the original intent. It is important to be clear about context and details relevant to the user.

Visually well-defined objects, meaning something with a lot of photos on the internet are preferable, such as terms like “wizard”, “rockstar”, or “mountain”. However, feelings and mystical-like themes like “a sense of awe or “the shores of infinity” will also work as well as describing a style like a “cyberpunk wizard.”

A user can use artists’ names to get a unique style from Midjourney (e.g. “Temple by James Gurney”) or combine names for new styles (“Temple by James Gurney and Ross Tran”). Furthermore, it is possible to ask for a particular medium like watercolor.

It is advisable to write in positives, avoiding negatives like “a hat that’s not red.” The program also accepts specific composition like “an ultrawide shot of a queen” but too many small details may affect performance (e.g. “a monkey on roller skates juggling razor blades in a hurricane”).

The user should use singular nouns or specific numbers (“three cyberpunk wizards”) while avoiding concepts that involve a lot of extrapolation (“an optimistic vision of an augmented reality future”). Finally, it is possible to combine two well defined concepts (“cyberpunk shinto priest”) [12].

Subscription and private mode

Without subscription, the prompts submitted by the users will be visible to all other that are part of the server [10]. There are two subscription tiers, $10/month and $30/month which allow the user to submit text to the bot as a private Direct Message, receiving images as responses. The first tier gives access to 200 prompts and the second, unlimited access [1] [13].

To enable private mode, the user must type the command /private on a Midjourney’s Discord channel. He will then be redirected to the payment site. To revert back to the public mode, the user must type /public in one the public channels. After this, it will no longer be possible to Direct Message the Midjourney bot [10].

References

  1. 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 Claburn, T. (2022). David Holz, founder of AI art generator Midjourney, on the future of imaging. Retrieved from https://www.theregister.com/2022/08/01/david_holz_midjourney/
  2. Borji, A. (2022). How good are deep models in understanding the generated images? arXiv:2208.10760
  3. 3.0 3.1 3.2 3.3 3.4 Midjourney Documentation. Quick Start Guide. Retrieved from https://midjourney.gitbook.io/docs
  4. Florian, M-C. (2022). Sistemas de Inteligência Artificial como DALL-E ou Midjourney podem realizar tarefas criativas? Retrieved from https://www.archdaily.com.br/br/987358/sistemas-de-inteligencia-artificial-como-dall-e-ou-midjourney-podem-realizar-tarefas-criativas
  5. The Economist (2022). Huge “foundation models” are turbo charging AI progress. Retrieved from https://www.economist.com/interactive/briefing/2022/06/11/huge-foundation-models-are-turbo-charging-ai-progress
  6. Mac, P. (2022). Shane 54 is pioneering a new way to make music videos with AI technology. Retrieved from https://www.youredm.com/2022/06/01/shane-54-is-pioneering-a-new-way-to-make-music-videos-with-ai-technology/
  7. Twitter. Shane 54. Retrieved from https://twitter.com/shane54music
  8. 8.0 8.1 Gault, M. (2022). An AI-generated artwork won first place at a state fair fine arts competition, and artists are pissed. Retrieved from https://www.vice.com/en/article/bvmvqm/an-ai-generated-artwork-won-first-place-at-a-state-fair-fine-arts-competition-and-artists-are-pissed
  9. Designboom. Midjourney. Retrieved from https://www.designboom.com/tag/midjourney/
  10. 10.0 10.1 10.2 10.3 Harmeet, G. (2022). Midjourney - Start your journey. Retrieved from https://weirdwonderfulai.art/general/midjourney-start-your-journey/
  11. Twitter. Midjourney. Retrieved from https://twitter.com/midjourney
  12. Midjourney Documentation. Tips for text-prompts. Retrieved from https://midjourney.gitbook.io/docs/resource-links/guide-to-prompting
  13. Bastian, M. (2022). Midjourney: DALL-E competitor enters open beta https://the-decoder.com/midjourney-dall-e-competitor-enters-open-beta/