Midjourney, the AI-powered image generation platform, recently released an alpha-test of Midjourney Version 5 (MJ V5) to showcase its new neural architectures and aesthetic techniques.
With improvements in many key areas, the latest version promises to take image generation to new heights.
In this post, we’ll explore the improvements that have been made since MJ Version 4, and what we can expect from future releases.
One of the most obvious changes in Midjourney Version 5 is the level of realism.
The Midjourney team calls this “unopinionated”, which basically means that it’s less artistic out of the box.
Therefore, the images generated by V5 are suitable for use in a broader range of applications, including advertising, gaming, and filmmaking.
The V5 images generated are more natural-looking, with improved lighting and shadow effects with greater ease and less effort.
One of the major issues with Midjourney V4 was the quality of the hands generated.
They often looked malformed, which could make an otherwise perfect image look unprofessional.
The developers of V5 have largely fixed this issue, resulting in more realistic and natural-looking hands.
The resolution of images generated by Midjourney V5 is also higher than V4.
The latest version generates images with a resolution of 2x, resulting in sharper and more detailed results.
This improvement is much needed for many applications, including advertising, that require clear, high quality artwork.
Another significant improvement in Midjourney V5 is the higher level of detail in the images.
This is especially evident in images generated with hyper-realistic prompt.
Midjourney V5 is also more flexible and versatile than its predecessor.
While V4 was primarily designed for generating still images, V5 can be used to create animations and even entire scenes.
This opens up a whole new range of possibilities for designers and artists, as they can now create complex, dynamic compositions that were not possible with V4.
Improved AI Models
The latest version of Midjourney uses state-of-the-art AI algorithms that are more powerful and accurate than those used in V4.
This means that the images generated by V5 are more sophisticated and nuanced, with more detailed and nuanced textures, shapes, and lighting effects.
Additionally, the AI models in V5 are trained on a more extensive dataset, which means that they can generate more varied and realistic images.
Color Still an Issue
While Midjourney V5 shows significant improvements in the above areas, the issue of color still persists.
V4 struggled to understand colors, with the model giving more importance to one color over the others when more than one was mentioned in the prompt. Unfortunately, this issue still persists in V5.
In the meantime, users can try some tricks to get better results, such as adding a plus (+) between the color and object, or giving more importance to a particular color in the prompt.
These are temporary workarounds until the issue is fully resolved in future releases.