Who says new version, says, in fact, improvements at all levels, both in the quality of the images generated and in its understanding of the requests or the artistic styles used. But all this comes with a certain trade-off.
Artistic intelligence
The competitor of DALL.E and Stable Diffusion is stepping up a gear with a version 5 rich in changes. From today, the very large community won over to his cause can have a field day asking him to generate images that are greatly improved in quality.
The heart of this new version is on the side of much more photorealistic images, thus placing Midjourney almost on an equal footing with DALL.E 2 on this ground. The images thus gain greatly in quality and in details displayed in a coherent manner. Particular attention has been paid to the hands and eyes, one of the biggest weak points of its previous versions.
Another major point of this version 5: a better understanding of queries, with a much more elaborate lexicon. It will therefore be easier to describe an image and obtain a result closer to what we imagined than before. The lexical field of Midjourney’s art styles has also been expanded to provide more variety.
The ransom of success
Of course, such improvements involve a quid pro quo. Such is thus the case at the level of the resolution of the images. This moves to default dimensions of 1024 x 1024, so that means Midjourney will take a bit longer to generate these than in its previous iterations.
This poses a certain problem for users taking advantage of the tool for free, or for those who do not have the most expensive subscription. A victim of its own success, Midjourney now imposes a limit not on the number of images generated, but on the time taken to generate them. As the process has become slightly longer, the time allotted to users will therefore be reduced accordingly.
Still, the already promising tool becomes even more impressive in what it is capable of achieving. An evolution that will certainly not be to the taste of communities of artists, who will have even more difficulty distinguishing a work generated by AI from a work produced by a human being. Not to mention the problem of copyright, which still remains nebulous with regard to art assisted by artificial intelligence.
