The latest version of the popular artificial intelligence image generator, Midjourney v5, can create incredibly realistic visuals. Midjourney v5 uses state-of-the-art tools and a new neural architecture to generate even more aesthetically pleasing art and design than the previous version.
It took five months to complete the new edition of Midjourney v5, and yes, it can now draw hands perfectly! The team announced the news on the official Discord channel of the project.
What is Midjourney V5?
Midjourney version 5 represents the latest evolution of Midjourney’s text to image conversion algorithm, which is regularly updated and improved. The chronology so far is as follows:
- Version 1 : March 2022
- Version 2 : April 2022
- Version 3 : July 2022
- Version 4 : November 2022
- Version 5 : March 2023
As you can see, the early versions went through relatively quick iterations, followed by a longer interval before Midjourney V4. Version 4 represented a considerable advance in terms of realism, image cohesion and overall aesthetics. After that, version 5 was again long overdue.
So… what has changed?
How do the results of V5 compare to previous versions?
As you can see, version 4 of Midjourney was already a major step forward in terms of image consistency compared to what existed before. Version 5 is a more subtle improvement that is not as dramatic. However, the level of detail and consistency is generally higher, and V5 tends to lean toward photographic images without other descriptors to guide it.
This fifth edition is a significant improvement, especially when it comes to creating famous pop culture characters. The difference is most noticeable when the neural network tries to represent a realistic scene, such as a picture, a landscape, or an interior space, like the example below.
Midjourney v5 : What’s new?
The release states that the update makes the following changes:
- A much wider range of styles and more responsive invitations.
- Significantly higher image quality (double the resolution) and improved dynamic range.
- More detailed images with more likely correct detail and less unwanted text.
- Improved performance with image invitations.
- Support for the –tile argument for seamless tiling (experimental).
- Support for –ar aspect ratios greater than 2:1 (experimental).
- Support for –iw argument to weight image prompts over text prompts.
What do they mean?
Stylistic: The aesthetic range of Midjourney v5 is wider than v4 and understands the users’ requests with more precision. The mention of artists’ names or movements has become more important in recent years. If no style is specified, the neural network will use its default setting to create a basic photorealistic output.
Quality: The resolution is now twice as high and the dynamic range is wider. Now high quality images can be created instantly. In version 4, the request was processed while the user waited for each image to be enlarged by hand. Now, by clicking the “Enlarge” button, a larger version of the image is immediately loaded. *
Detail: When it comes to small details, the neural network excels. For example, it produces five fingers on each hand and does not add hundreds of uniform teeth. The developers also claimed to have “significantly” improved face detail.
The latest version of Midjourney allows you to create photorealistic visuals, with a higher level of graphic detail than expected. However, the creators caution against the effectiveness of two-word prompts and recommend longer, specific prompts. Although the development phase of the fifth generation is still underway, it is already available to the general public, the developers warn that significant changes could still occur before the final release.
How to use Midjourney v5?
Currently, the default version of the Midjourney neural network is v4.
You have two options to activate the fifth one:
Simply type /settings into Discord via the Midjourney bot, choose the MJ option, and press Enter to apply the changes. Choose MJ version 5 in the menu that appears, then click Apply.
Or add –v 5 to your prompt when generating an image with the bot