Artificial intelligence image generator MidJourney has released version 6 of its popular tool, allowing for longer prompts, greater levels of detail and even adding text to a photo.
MidJourney first entered open beta in July 2022 and has made rapid and impressive progress in the quality and accuracy of images with each subsequent release.
This release comes at a time when MidJourney is under increasing pressure from competitors including OpenAI’s DALL-E 3 which can generate text. Similarly, free tools from Meta and services like Leonardo also offer increasingly more control over image generation.
What is new in MidJourney v6?
The latest release of MidJourney includes a longer prompt length, more granular control over color and shading, text and the ability to fine-tune the output through a conversation with MidJourney in much the same way ChatGPT lets you refine DALL-E images.
The latest update has been in development for some time, with select groups of subscribers testing and comparing the output to improve the overall quality.
One of the most striking updates is in the way it interprets and understands prompts. For example, it can now understand nuances in both punctuation and grammar, telling the difference between eating, shooting and leaving.
Where can you access MidJourney v6?
Like all previous versions of MidJourney, version six is available through Discord. Once you’ve paid the subscription you’ll get access to the bot in messages where you can /imagine the creation of your choosing.
To access version six you’ll need to either select it as the default model in settings or append –v 6 to the end of any prompt. This will ensure it uses the correct model and still allow you to continue using version 5.2 by default.
MidJourney has started opening access to a web version for users that have generated more than 10,000 pictures. This includes the ability to change settings through different inputs rather than relying on the AI model to understand a single text prompt.
What do MidJourney v6 images look like?
To test out the latest version (which MidJourney stresses is still in alpha release and review mode) I set it a series of prompts and compared the output to the same prompt in v5.2.
Testing out the text
(Image: © AI generated with Midjourney)
For the first prompt, I asked MidJourney v6 to create a poster for a movie called Moon Pups but it failed to add the text to the image.
I decided to be more explicit and tell it to include specific words, placing them in quotation marks for emphasis. That solved the problem, although MidJourney v6 has the same issue as DALL-E with spelling.
Side-by-side comparison
(Image: © AI generated with Midjourney)
Next up I had MidJourney v5.2 (left) and MidJourney v6 (right) each generate an image from the same prompt to compare. I found that while v6 was richer in detail, v5.2 benefited from months of refinement and had more character.
The prompt: “wine taster peering through a wine glass.”
Real people
(Image: © AI generated with Midjourney)
Finally, I had both v5.2 (left) and v6 (right) create an image of a real person. Some AI models refuse to produce a picture of a living human including DALL-E but MidJourney doesn’t suffer from those same concerns.
I had both create a picture of Elon Musk riding a unicorn. Neither was given any prompt refinement or further instructions. Version 5.2 created a cartoon-like depiction whereas version 6 looks like a photograph.