MidJourney v5 represents a significant qualitative leap in image generation with this AI. It’s too early to evaluate the version, as it is currently in alpha state, meaning it’s in a testing phase. It’s interesting that the people behind MidJourney decided to release it. Normally, an alpha version wouldn’t be released to the public and would be tested by the development team. Once that’s done, it should move to beta status and then could be tested by end users. In any case, we have v5 available with all the drawbacks that come with a version still in development. However, there is a clear improvement over previous versions. The photographic quality, for example, has greatly improved. Almost authentic photographs are achieved. Issues such as the generation of anatomically correct hands are being fixed rapidly. Sometimes a hand with extra fingers or misplaced fingers still appears, but these cases are increasingly rare. The duplication of legs and arms only occurs in some cases. Eyes no longer tend to appear misplaced or with deformed pupils. In this aspect, the improvement is significant.
On the other hand, prompts are not interpreted the same way as before. There are theories about them advocating for long and very descriptive prompts over short and keyword-built prompts. Experimentation will teach us the best technique for our work. For example, if we want to have a catalog of emotions, it used to be enough to give a simple prompt like ‘Anger’ or ‘Elation’, and we would get an image representing that emotion. Now it’s more complex. MidJourney wants us to provide more details, like ‘/imagine prompt: Color photography of Human emotion:: Two people, a man and a woman feeling Guilt’. It’s a change in the way of thinking about prompts.
In the illustration section, there is a noticeable change. Just mentioning illustrators is no longer enough. MidJourney tends to ‘photograph’ everything, so the result might not be satisfactory for those looking for the texture of oil or watercolor. We need to indicate it more explicitly. And it does it well. The imperfections typical of pictorial art may not be as well represented unless, as I said, we specify what we are looking for. v5 has other changes, like the recovery of the –tile or –iw parameter, and for the time being, there’s no upscale since the image grid is already generated at the highest final quality, that is, 1024×1024. Just use the Un button to get the separate image in that quality (the same effect can be achieved in the full grid if we use the envelope emoticon, as when we want to get the seed back). Ratios are now completely unrestricted. By default, you can use 1:1, portrait or landscape options such as 9:16 or 16:9, and even better, extreme ratios for banners and more, like 20:1, 5:30, or anything else you desire.
In summary and very briefly, v5 is undoubtedly a qualitative leap. Let’s wait and see the final version and the upcoming v6 already in development. The competition doesn’t have it easy. It’s true that MidJourney has shortcomings, such as the absence of user-trainable models, the possibility of using the app in local mode, the lack of options like inpainting and outpainting, not having a way to request a pose or position for the characters, but we’re on the right track.
