Midjourney Version 5 vs. Version 6
Back in March of 2023, one of my colleagues, Luther Andal, wrote a blog post showing images created in Midjourney version 5 alpha. One of those images, a Sioux chief, struck me at the time as very nice and mostly realistic. Now, more than a year later, I decided to use the same prompt to compare a similar image from Midjourney version 6 to see if the newer generated images appear more realistic. Let’s test it: Midjourney version 5 vs. version 6.
The original prompt follows: a close up of a Sioux tribe chief wearing traditional headdress and clothing. I used the original author’s prompt exactly as he wrote it, but this time I used Midjourney 6 alpha. Here are the results.
Image 1: Midjourney Version 5
Image 2: Midjourney Version 6
Comparing Images from Midjourney Version 5 vs Version 6
At first glance, the two images look quite realistic. When I first saw the one generated in version 5 over a year ago, I thought it was amazing. Now, though, I think the newer image is superior to the older one.
First, notice how the folks in the skin in the version 6 image appear very natural, while the ones in the version 5 image seem slightly exaggerated. This difference is minor, but I see it.
Second, the skin is quite different. The older image has skin that looks a bit more shiny. At the time, I thought it looked amazing. However, in comparison, the newer generated image has skin that’s way more realistic, including the lighting, shine, and texture.
Next I noticed the difference in skin color, but I think that’s probably a fairly random result, given the range of actual skin tones in humans, even ones from a specific group. Initially I thought the v5 image looked great, but now I’m more drawn to the color of version 6, which appears to me to be more accurate. I will admit I don’t know which is closer to reality, though both may be equally realistic or not.
Finally, somewhat unrelated to the reality of the faces, is he headdress and apparel. They’re both striking, but the vibrancy and clarity of the version 6 image is outstanding. I wonder if the designs are actually related to Sioux culture or if Midjourney is mixing them with other classifications of Native Americans. Accuracy of a wide array of skin tones is important to present humans as they truly appear, but so are cultural attributes such as apparel. A Sioux-centric prompt should generate an image of a person who appears like a Sioux, including what he/she wears.
Conclusion
I am stunned by the level of quality Midjourney produces, including how realistic some of its images are. It continues to improve, resulting in nearly photo-realistic images that can fool viewers, assuming they haven’t already. It’s a fantastic application that I’m sure will continue to generate images in an even more deep and wide ways, both creatively and realistically.
Note
One fundamental problem of any AI image generator is not just the lack of racial or cultural accuracy, but he inability of the viewer to spot an inaccuracy. For example, an image of a city in one country might include a building that appears to be a specific one from a city in a neighboring country. The viewer, in this case me, might be utterly unaware of the discrepancy and remain ignorant of any inaccuracy, since the architecture may appear to be consistent among all the buildings in the scene. I assume Midjourney and other AI image generators will continue to produce unbelievably accurate images over the coming years.
Contact the Author
Please feel free to contact me. If you have a useful question or comment, I’ll add it here with my thoughts as appropriate, though I will never post your name or email address.
Bill Adams
[email protected]
If you would like to speak to someone at Streamline about a project, please contact us any time.
