It is very easy in Midjourney to create a prompt using a well-known artist’s name like Artgerm or Moebius, and the results frequently lack the original's essence and seldom capture the true spirit behind the artist's work.
It is akin to using a filter on a photograph—the underlying image might change in appearance, but its essence remains unchanged.
For instance, if you add by Pablo Picasso to your Midjourney prompt, what period of Picasso’s work would Midjourney imagine?
All of Picasso’s seven distinct periods are instantly recognizable as being painted by him, but each is unique in style. Here are just four examples for comparison:
.jpg)
.jpg)
.jpg)
.jpg)
Midjourney just doesn’t recognize the subtleties of various periods of the great master’s work.
In the same way, different artists from different periods do sometimes exhibit similar styles. Take the artwork of the American illustrator N. C. Wyeth, for instance. Other artists like Howard Pyle, Joseph Lorusso, and Jeffrey Catherine Jones all have certain aspects of their art that look like NC Wyeth’s artwork, even though some of them come from completely different artistic periods and have unique styles.
All that made me wonder: how can we direct Midjourney toward an original style without relying on its ability to mimic the original artist’s work? And how can we construct and create our own styles without directly referencing the artist’s names in the prompts?
Enter Mimetic Style Emulation
The method that I have developed and refer to as Mimetic Style Emulation is not about copying but about understanding.
It is about diving deep into what makes an artist's work resonate and then using that understanding to create something new, inspired by the original style. It is a tribute, not a replication.
Ever since the early 1980s, at the very beginning of my artistic career, I have always admired the work of the fantasy artist Jeffrey Jones. What made his work so distinctive? Was it the ethereal quality of his landscapes? The soft, muted color palettes? The way he portrayed light and shadow? Or perhaps the melancholic expressions of his characters?
After Midjourney appeared on the art scene a little over a year ago, in July 2022, I could begin creating some “Jeffrey Jones” artwork by using a prompt like this one:
This could be the result, and it certainly looks somewhat like Jeffrey Jones’ work. But what other words could I use to get the same look without using Jeffrey Jones’ name in the prompt? In other words, how would you describe his artwork to someone who has never seen it?
While I was struggling to emulate the style of Jeffrey Jones, I encountered Christian Heidorn’s YouTube video and blog post↗︎, where he outlines a totally unique method that uses ChatGPT to emulate a particular artist’s style.
I tried Christian’s method myself but only had limited success. The best prompt that ChatGPT could create, using the words Fantasy art, Mythical creatures, and Imaginative realms, produced several rather underwhelming results, not looking at all like Jeffrey Jones’ style!
While Christian Heidorn's approach was innovative, for me, it only highlighted the fundamental differences between textual and visual mediums.
ChatGPT, being a text-based model, thrives on detailed, descriptive prompts. It's designed to understand context, nuance, and the intricacies of language. On the other hand, Midjourney's algorithm is more abstract, focusing on the decomposition of prompts into tokens to generate visual art.
To illustrate this, here is a good old human-made prompt: astronaut lost on alien planet surrounded with red extraterrestrial bioluminescent flora. Then I talked to ChatGPT:
Please write a prompt for a Text-to-Image AI featuring astronaut lost on alien planet surrounded with red extraterrestrial bioluminescent flora.
Certainly! Here is the prompt: “Render a lone astronaut discovering an alien garden under a bioluminescent sky, mixing surrealism with high realism, focusing on contrasting alien flora”.
The final step was to “feed” the initial image to Midjourney’s /describe:
Human:
ChatGPT:
/describe:
This distinction is crucial! When we use ChatGPT, we are essentially painting a picture with words, providing as much detail as possible to guide the model towards our desired outcome.
However, with Midjourney, the process is more about distillation. It's about reducing a concept or idea to its most basic elements and then allowing the algorithm to interpret and recreate based on those pieces.
Human:
ChatGPT:
/describe:
This is where I believe many, including myself, initially stumble when trying to emulate a particular artist's style using Midjourney. We're so conditioned to think in terms of detailed descriptions, especially when using tools like ChatGPT, that we forget the power of simplicity and abstraction.
To truly harness the potential of Midjourney, one must think in terms of tokens—the basic building blocks of an idea. Instead of trying to describe every nuance of Jeffrey Jones' style, I thought it might be more effective to identify the core elements or themes that define his work visually and use those as prompts.
It was then that I discovered the /describe command that was added to Midjourney shortly after Version 5 was released earlier this year, and this was the breakthrough that I was looking for.
%20web%20OPT.jpg)
Read our In-depth guide on Midjourney's /describe →
One of the biggest advantages of the /describe command is its ability to "reverse-engineer" an image. It is very beneficial when you have an image with a style or “look” that you wish to emulate. But the major benefit that I personally obtained from /describe, was that it helped me discover new useful tokens to use in my prompts.
Searching for and categorizing these special “token-words” that Midjourney generates would have been a time-consuming activity… if not for ChatGPT doing all the hard work for us humans!
As you can see, it's an effective method for adjusting the 'sharpness' in your image and taming styles (whether tuned or not) that tend to be excessively detailed.
Stop
For this test, I selected eight style directions from the set, aiming for more illustrative images, even though most were quite far from Miyazaki’s distinctive style:
Here are the results of applying this surprisingly consistent style to some of the benchmark prompts:
Mimetic Style Emulation: step by step
Firstly, I found an image on the Internet of artwork that I felt was typical of Jeffrey Jones’ work.
Secondly, I uploaded it into Midjourney using the /describe command to create four possible prompts. Re-rolling it three or four times gave me a broader variety of prompts to work with.
(I found 16 was a good manageable number that would give me enough variety in prompts)
Here are just some of the results:
With the 16 prompts saved in a text file, I then began “training” ChatGPT to analyze the prompts and to do the following:
You are my personal assistant and will help me analyze the prompts that the Text-to-Image AI program Midjourney produces with its /describe command. This task is going to be called “Prompt Breakdown.” I will provide the prompts in the following format. Please store and remember these:
“a portrait of someone in a tree, in the style of mythological paintings, implied movement, paleocore, epic fantasy scenes, art that plays with scale, , light yellow and dark brown --ar 34:49”
1. Separate description (subject, context, backdrop, etc.) from the style modifiers (artistic techniques, references to artists, medium, etc.).
2. Retain the words ”In the style of“ in the breakdown.
3. Create a table with columns for Description and Styles, and rows for each prompt (numbered 1-16).
4. Add repeated terms at the end of the table under the title ”Repeated Terms“.
5. Once approved, create a downloadable .CSV file with the table.
I redacted all the artist’s names, like Frank Frazetta, and then returned to Midjourney to feed it with the analyzed prompt containing combinations of the repeated terms. I also added a specific aspect ratio, and used a fixed seed number, so I could compare the images using different token-words, but retain the same basic image.
I didn’t expect to get the perfect image after the first roll. Sometime several re-rolls were needed, and using a fixed seed allowed me to easily compare any tweaks that I made:
I also remembered that the tokens at the end of the prompt are not as important as the ones at the front (coming right after the description), so I experimented with changing the order of the tokens, even removing some of them, to see if they would influence the generated image.
For instance, I put art of the upper paleolithic first and removed junglecore and twisted branches. I immediately got better results:
I found some interesting token-words (that were not repeated) in the CSV, and began to experiment with them to see if their inclusion would change the images in any way.
Here are some of the words I identified this way:
loose handling of paint, post-painterly, trompe-l'œil illusionistic detail, oil on panel, varied brushwork, heavy use of palette knives, verdaccio underpainting, underpainting with oil washes, made of mist.
Using different combinations of these tokens that the /describe command had produced, the resulting images came very close to a style similar to that of Jeffrey Catherine Jones:
As you can see, it's an effective method for adjusting the 'sharpness' in your image and taming styles (whether tuned or not) that tend to be excessively detailed.
Stop
For this test, I selected eight style directions from the set, aiming for more illustrative images, even though most were quite far from Miyazaki’s distinctive style:
Here are the results of applying this surprisingly consistent style to some of the benchmark prompts:
Insight!
Aspect ratio (--ar or --aspect) might dramatically influence the outcome of your prompt.. Consider the composition of your image and what format is best for it: horizontal (landscape), vertical (portrait), square, etc. You can consider the aspect ratio of your original “training” or “seed.” It will often be a good starting point!
Conclusion
For budding artists or even seasoned professionals looking to expand their horizons with Midjourney and similar Text-to-Image AI programs, I recommend experimenting with this technique.
You can use this method of Mimetic Style Emulation to study the masters, understand their visual essence, and then let that understanding guide your own MidJourney creations. The results can be both surprising and deeply satisfying, as you can see below.
I will show the original Jeffrey Jones artwork that I used as inspiration, with the results afterward:
Insight!
You might notice that I add the --style raw parameter (or alternatively) --stylize 0. This is important to get accurate results and to steer Midjourney away from inserting its own aesthetics into the image whose prompt you have so carefully crafted. --style raw or --stylize 0 essentially switches off Midjourney’s internal aesthetic and forces it to stick more closely to your Mimetic Style Emulation prompt.
In conclusion, while technology like Midjourney does offer a shortcut to mimicking styles, true artistic growth comes from understanding and internalizing the essence of the artists we admire. The Midjourney and ChatGPT tools have the ability to quantify these special elements. By using this sort of analytical method, you have the potential to not only pay homage to their genius but also pave the way for your own unique artistic journey.
— David Wiles (aka hamfast)
If you like our guides and studies, please, consider supporting us. It's thanks to our Patrons that we are able to maintain and develop Midlibrary, create better educational content, and keep it free for all!
Support Midlibrary on Patreon! →
ⓘ New styles on every page!
All samples are produced by Midlibrary team using Midjourney AI (if not stated otherwise). Naturally, they are not representative of real artists' works/real-world prototypes.
Ver. 2.9.1
♡
We do our best to keep this website running as smoothly as possible. However, stuff happens, and we thank you for letting us know!
We're here to help! If you're unable to resolve the issue, please, contact us.
We regularly publish new Midjourney Guides, compile new Style Tops, update the website, and have fun! Want to be the first to get Midlibrary news? Subscribe to our newsletter and never miss a thing!
If you're unable to resolve the issue, please, contact us.
Personal Libraries are available to our Patreon Community
Learn more about the benefits of supporting us by becoming Midlibrary Patron—and start your Personal Library ↗︎
You have just become a Patron, and cannot log in?
Please, allow our team some time (usually not more than 24 hours) to set up your Personal Library.
You may be using different emails for your Patreon and Discord accounts. If that is the case, please, send your Discord email to [email protected].
If the issue perists, or you didn't get a response to your email, please, inform us via Bug Report form
We are currently updating the Personal Libraires' infrastructure
In the nearest future, it will allow you to access your Collections much quicker, add covers to them, tag the styles you save to quickly find them, and—most importantly—save your --sref (numerical) styles!
However, at the moment, logging in to your Library is unavailable. We apologize for the inconvenience. If you are a Midlibrary Patron, please, check this Patreon post ↗︎ for Personal Libraries status updates.
To start creating Collections and save favorite styles:
Log in with Discord →
Learn more about Personal Style Libraries, saving favorite styles, and organizing them into Collections.
Learn more about supporting Midlibrary and the benefits of joining our Patreon community →