In this experimentation of AI tools for creatives, I'm going to ask Midjourney to describe to me using it's own AI model what the famous image by Sandro Botticelli is, and then re-run that description to see how closely we can replicate the mastery of this artist.
About the AI Tool (Midjourney)
Midjourney has emerged as a favored AI image creation tool within the artistic community. Having debuted only a few years back, it has swiftly established itself as an essential component of a creator's toolkit. Among the functionalities we'll be leveraging are Midjourney's "Describe" and "Imagine" commands.
Midjourney /describe
The describe command is used for two essential purposes, explaining the contents of a visual and determining the dimensions of a visual. According to Midjourney's website, "The /describe command allows you to upload an image and generate four possible prompts based on that image. Use the /describe command to explore new vocabulary and aesthetic movements."
Midjourney /image
The imagine command is where all the magic happens! Imagine is the way we ask Midjourney to create the images we are looking for. According to Midjourney's website, The /imagine command generates a unique image from a short text description (known as a Prompt).
Replication Test Artist and Work: Sandro Botticelli's Birth of Venus
Botticelli's Ai Artwork Description
The first thing we're going to do is take our sample image (our valued piece of art history!) and ask our AI tool, Midjourney, to give us what it think is an accurate written description of the contents of the image.
Interestingly enough, Midjourney did use the name Sandro Botticelli in one of its descriptions! We also do see a healthy amount of reference to the birth of venus in every single suggested prompt description. Here are the ai prompt descriptions from Midjourney:
Description 1
the birth of venus by turin, in the style of feminine empowerment, elba damast, anemoiacore --ar 50:31
(a note: the -- lines preface commands in midjourneys prompt formatting. in this case --ar 50:31 is the dimension of the image! "ar" stands for aspect ratio)
Description 2
the birth of venus, in the style of feminine empowerment, polished metamorphosis, florentine renaissance, hatecore, natural beauty, marine painter, symbolic nature --ar 50:31
Description 3
birth of venus, in the style of feminine empowerment, polished metamorphosis, feminine body, hatecore, florentine renaissance, natural beauty, sfumato --ar 50:31
Description 4
a painting is shown that features a painting of the birth of venus, in the style of fanciful elements, graceful forms, anemoiacore, golden palette, basil gogos, pegi nicol macleod, sandro botticelli --ar 50:31
I also noticed that some of the terms are not quite what I would use to describe this body of work, and that Midjourney has a weird trend of making up it's own "styles" by adding the word "core" to the end of a word. We see things like "hatecore" which I am very baffled by, and "anemoiacore".
Sandro Botticelli Inspired Still Life Image Generations
How much value is in the artists name?
The first thing to test is an understanding of how the language model responds to the usage of the artists name in the prompt. In this case Sandro Botticelli was used in description 4. We only have one test to see the visual weight of this author, but one thing to consider is that it also looks like Midjourney put other artists as inspiration as well. Which one will take priority?
Description 4 from Midjourney:
a painting is shown that features a painting of the birth of venus, in the style of fanciful elements, graceful forms, anemoiacore, golden palette, basil gogos, pegi nicol macleod, sandro botticelli --ar 50:31
Your immediate reaction is probably just as unexpected, surprised, and curious as mine. I see some inspiration of a warped version of the Birth of Venus in the first image (top left) and the third image (bottom left). I can see some resemblance of similarity in the second one (top right) only through the fact that we have considered the sea shell in the image, and the last one (bottom right) is just plain wrong.
The more I thought about it the less unexpected it was though, because we did have some additional artist inspiration. It looks more like the stylistic words of fanciful, graceful, and then Midjourney's v 6.0 styling outweigh the rest of the parameters.
For the purpose of this experiment, let's compare side by side the artists work against the most favored Midjourney rendering in the grid of four.
In this first grouping, I really really didn't even think there was one close enough to compare, it feels disrespectful to the artist to be this far off base in style, but nonetheless we shall review! I chose the first image in the grid becuase it showcased a similar amount of movement and figures to the original Botticelli piece. The AI, however, has taken many liberties to adjust not only the color palette but the entire application of the medium to convey the desired visual.
I would not say this is a successful attempt to replicate Botticelli's work.
What does it render without the artist name?
Now that we've taken a look at the description rendering that included the name of the artist, lets see if focusing more on the subject matter and less on the artist himself changes anything. Descriptions 1, 2, and 3 can shed some light on the importance of the different terminology used within an AI prompt. Even if we lack key distinguishing nouns, if we provide the right set of adjectives we can see if that will give us a more desirable result.
Let's take a look!
Description 1 from Midjourney:
the birth of venus by turin, in the style of feminine empowerment, elba damast, anemoiacore --ar 50:31
The first thing I noticed when running description 1 is that we actually have an inaccurate author attached to this prompt. "Birth of Venus by turin" likely has some insight into why we got pretty close with the subject matter but less so on application. Narrowing it down I was stuck between image 1 and image 3 in the rendered grid. Ultimately, based purely on the lighting I decided that image 3 felt the most similar to our desired visual representation. This one is just absolute chaos to me with all the little children added to the scene, I wonder if "turin" had something to do with that?
Description 2 from Midjourney:
the birth of venus, in the style of feminine empowerment, polished metamorphosis, florentine renaissance, hatecore, natural beauty, marine painter, symbolic nature --ar 50:31
The AI generated images from description 2 are also a challenge. We see terms like "feminine empowerment" and "renaissance" that I think help tie into the style of the figures in the results, but we still see other terms changing our results a bit. While the first and third images I really, really love, it is clear that they are not accurate representations of Botticelli's style. When we consider the medium and lighting, I think image 2 gives us the best representation of embodying the traits of a Botticelli work.
Description 3 from Midjourney:
birth of venus, in the style of feminine empowerment, polished metamorphosis, feminine body, hatecore, florentine renaissance, natural beauty, sfumato --ar 50:31
Our final rendering completely omits all mention of artists. We are purely relying on the title of a famous piece of art and then other stylistic attributes to reimagine that image. Based on how this prompt is structured, the results are pretty expected. It feels very inspired by the subject matter of the Birth of Venus, and I feel like we have gotten the closest in application of that visual with image 3 in this grid. There are definitely some interesting adaptations and 3D effects rendered here as well, but the creative deviation of this image is very expected based on the way the prompt was structured.
Can We Replicate History with AI?
If there is one key takeaway that these experiments have shown us, its that Midjourney does have references of famous artists and their works in it's model. It was able to pin point the exact name of the piece and even give us the author in one of the prompts. The unfortunate byproduct of adding in other terminology clearly shows an undesirable deviation from our artist's specific style.
After going through this initial test, it is easy to see how we can emulate the styles of famous artists through language if we test our prompt style and put together the right string of words. It's also easy to see how quickly that can become wildly inaccurate. (Sometimes in a really fun and unexpected way, sometimes in a stressful way.)
As one last and final curious experiment with Midjourney, I simply entered in "/imagine Sandro Botticelli's Birth of Venus" to see if, without any context at all, Midjourney would replicate the art in question better or worse than when it was given the image to describe first.
As I had anticipated, combining just the artwork with the author name was a much more expected set of results. Are they perfect? Absolutely not, but we finally have a more accurate application of medium, tonal matching, and figure rendering.
Let's Keep Experimenting With AI and Art History!
Stay tuned for Part 3, where we'll explore the work of another famous artist and see if Midjourney can replicate it! Who will it be next time?