The new version of the world’s best AI image generator is here at last! Equipped with the latest AI model, Midjourney v6 (Alpha), this version comes with more than just a quality upgrade. What’s new in the Midjourney AI art generator, and how can you get the most out of it?
What is Midjourney v6?
Midjourney, a small independent research lab of just 11 full-time employees, has been developing some of the best generative AI image models of 2023. With the goal of “exploring new mediums of thought and expanding the imaginative powers of the human species”, this ambitious studio has just secured the crown once more.
CEO David Holz announced the latest update via the Midjourney Discord server.
As you can imagine, there’s hot competition for the #1 spot in AI image generators. OpenAI has its own model, DALL-E 3 (Stylized DALL·E), which challenged the top position earlier this year due to its ability to produce clean and legible text. This ability, known as text-in-image, is the rendering of alphanumeric characters in situ as part of an AI image generation.
It’s a completely different process from text editors such as Microsoft Word and Google Docs, and much more difficult to get right. Whereas the former simply refers to a library of said characters, in a typeface that comes pre-designed, a generative AI model must recreate the typeface consistently and from scratch every time using an algorithm that is largely designed for producing aesthetically pleasing, but fungible, images.
In other words, an image of a mountain is an image of a mountain. An image of a street is an image of a street. That same can be said of a face, a car, or a cat. There’s a lot of room for variation when we aren’t asking for a specific person’s face, but only a face that we recognize as human. We don’t have the same specificity or consistency expectations of a mountain as we do of any given letter of the alphabet. It must be a specific shape, otherwise we won’t recognize it, and then it loses its entire purpose! This is what AI image generators have struggled with… until now.
Essential AI Tools
What’s new in Midjourney v6?
The full list of new and improved functionality with the alpha version of v6 (Update 21.12.23) includes:
- The prompt length has been increased to 1,985 characters. This includes spaces and the prompt prefix “/imagine:”.
- It’s better at conceptualizing object permanence and physical space.
- Text-in-image is 40% accurate, 40% of the time in our testing at PC Guide (better than before).
- Improved natural language processing (NLP) results in a better prompt understanding of punctuation and grammar.
- Chat to Midjourney like ChatGPT. The conversational style with which you can prompt has been altered because it’s better at ignoring filler words and prioritizing ‘useful’ words.
- It better understands punctuation and grammar.
- You may use linguistic negatives such as “no mustache” and it will adhere to the negative, unlike previous models.
Midjourney v6 prompting tip
I found the character limit for v6 Alpha by asking ChatGPT to generate a string of text exactly 2,000 characters long. This didn’t fit, so I worked my way backwards until I reached the figure 1,985. To make the most of this space (which is already more than enough), use “/settings” in Discord to enable the v6 model. This way, you wont need to use characters to specify the model as a suffix in your prompt.
How to enable Midjourney v6
To enable Midjourney v6, you’ll need a subscription to the service. With a basic plan costing just $10/month, you’ll receive 3.3 hours of “fast” GPU time each month. Try out the latest updates of Midjourney in your image prompts, with improved world knowledge and higher resolution than v5 – just be sure to include the — —v6 suffix! (Two hyphens, followed by v6, with no spaces)
To provide more clarity around text-in-image generation, Midjourney can now write legible words but it will take a few tries. The neon sign shown in this article was, impressively, a first attempt with that art style, but it still struggled at ’17th-century parchment’ on its eighth attempt.
Summary
Midjourney v6 is the latest version of the AI image generator. With enhancements including higher image resolution, up to a maximum of 2048×2048, smarter natural language processing (NLP), and more realistic human faces, hand generation, and even text, this represents the most impressive iteration yet.
Midjourney fans who contributed to the first rating party, organized by Midjourney founder David Holz, helped to shape the latest Midjourney version by labeling ‘good’ images, as well as ‘boring’ or ‘junk’ ones. This community-sourced reinforcement learning with labeled data is something Holz considers a “very important part of [the firms] process”. Enthusiasts of this artificial intelligence software instantly took to social media to share the significant improvements of its new capabilities.
