AIs image generators are a hit

A.I. tools like DALL-E which can generate images from a text input, are getting very popular for those who create content, but they are also worrying for artists who are afraid to be replaced by an A.I. However, new tools have been coming out since the spread of DALL-E, and Midjourney seems to be its first competitor.

According to Vice, Midjourney has launched its closed beta, allowing anyone to create an account and create fantastical renderings, allowing some “free” generations. The system, which matches DALL-E in its capacity to produce amazing and occasionally unsettlingly lifelike renderings, underwent a stress test as a result of the influx of new users.

Compared to DALL-E, Midjourney appears to have a distinct talent for setting up scenes, particularly fantasy and dystopian sci-fi settings with dramatic lighting that resembles realistic concept art from a video game. It’s also incredibly effective at creating bizarre blendings that imitate many artistic movements.

The test gave way to a fully-open beta, allowing anyone to sign up and join the project’s Discord channel. The beta is operating entirely through Discord, with users typing their prompts directly into the chat interface and receiving messages from a bot that shows their generation rendering in real-time. Users can then choose to upscale and enhance an image from each set of generations or create more variations from the same prompt.

midjourney
midjourney
midjourney
midjourney

However, each user is only allowed a certain number of generations during this “free trial” period before the bot asks them to subscribe. For non-commercial use, the cheapest plan costs $10 per month and offers 200 photos; the most expensive plan costs $30 per month and offers infinite generations. (The creators of DALL-E recently polled its beta testers on potential price points, and they also intend to charge for access to their AI tool).

The requirement that anyone using generated photos in “anything linked to blockchain technologies” pay a 20% royalty on any earnings over $20,000 per month is another way that Midjourney uses to deter people from minting NFTs.

While the creators of DALL-E have made an effort to reduce some of the biases in the training process that are inherent to these models, including violent and sexual content, Midjourney hasn’t made any disclosures regarding the datasets and techniques used to train its AI tool and doesn’t appear to have any explicit content protections beyond automatically blocking specific keywords. However, upstream censorship could lead to a limited use of the tool, even when there aren’t malicious purposes but only artistic ones. It is also true, however, that being able to produce extremely realistic material with violent content involving real people would be detrimental, but the same content in a clearly unrealistic style would certainly be less harmful.

The “Content and Moderation” section of the Midjourny user guide gives users instructions on how to avoid making visually shocking or disturbing content, including adult material and gore, as well as how to avoid creating images or using text prompts that are inherently disrespectful, aggressive, or otherwise abusive. Aside from objectionable photos of public figures, the regulations also prohibit material that may be perceived as racist, homophobic, unsettling, or in some manner disparaging to a community.

These AI tools are changing the way creators and artists approach art and content. Will creators mass-use these tools? Or will they adopt them just as a help? We already have texts generated through AI tools, but in the future, we could see content completely generated by Artificial Intelligence. AIs will be able to create music, videos, texts, and images, and in part, they already do it. So, will artists have to be more game-changing to be original? Or will they employ this technology to create a new form of art?

Banner

Here’s a brief timeline of the evolution of Midjourney features:

  • Midjourney V2: Released on April 12, 2022, this version introduced upscaling and variation buttons along with a new model. The Midjourney team honed down on a concrete pricing plan and switched to a paid beta.
  • Midjourney V3: Released on July 25, 2022, this version introduced the --stylize and --quality parameters. Low stylization values produce images that closely match the prompt but are less artistic. High stylization values create images that are very artistic but less connected to the prompt.--quality instead, changes how much time is spent generating an image. Higher-quality settings take longer to process and produce more details. Higher values also mean more GPU minutes are used per job. The quality setting does not impact resolution.
  • Midjourney V4: Released on November 5, 2022, this version brought an unprecedented level of quality, far beyond what any existing Stable Diffusion model could produce. This model featured an entirely new codebase and brand-new AI architecture designed by Midjourney and trained on the new Midjourney AI supercluster. This version increased knowledge of creatures, places, and objects compared to previous models. It also has very high Coherency and excels with Image Prompts.
  • Midjourney V5: Released on March 15, 2023, this version continues the quality and versatility upgrades of the previous version. This version produces images that closely match the prompt but may require longer prompts to achieve your desired aesthetic.
  • Midjourney V5.1: Released on May 3, 2023. This model has a stronger default aesthetic than earlier versions, making it easier to use with simple text prompts. It also has high Coherency, excels at accurately interpreting natural language prompts, produces fewer unwanted artifacts and borders, has increased image sharpness, and supports advanced features like repeating patterns with--tile.
  • Midjourney V5.2: Released on June 23, 2023. From this version, you can fine-tune results with the --style parameter to reduce the Midjourney default aesthetic. This model produces more detailed, sharper results with better colors, contrast, and compositions. It also has a slightly better understanding of prompts than earlier models and is more responsive to the full range of the --stylize parameters.
  • Midjourney V6: Released on December 21, 2023, after 9 months of development. It brought outstanding improvements in image quality and encouraged simpler prompts. Here are the major changes:
    • Upscales are now 2x faster (and use up 2x fewer GPU minutes)
    • Improved aesthetics, coherence, prompt adherence, and image quality.
    • Improved text rendering (you must put text inside “quotations” in your prompt)
    • Improved performance at high --stylize values.

Here’s the feature evolution of Dall-E:

  • DALL-E V2: Released in 2022, DALL-E 2 sought to generate more realistic images at high resolutions, combining concepts, attributes, and styles. To achieve this feat, DALL-E 2 improved the techniques used. For example, the DALL-E 2 generates higher-quality images using a stable diffusion model that integrates data from the Contrastive Language-Image Pre-Training (CLIP) model, which was trained on 400 million labeled images.
  • DALL-E V3: Released in November 2023, DALL-E 3 represents a significant step in AI-based art generation. It improves on many of the previous limitations possessed by its predecessors, DALL-E and DALL-E 2, as well as generating media more accurate to the prompt than Midjourney.