Imagine a machine that literally turns your ideas in to original works of art

Is AI ready to replace Medical Illustrators?

It is called Midjourney and by simply describing what you want to see, it can paint, draw, photograph, or visualize a sculpture in a matter of minutes. It is the first truly versatile AI art generator and even though it has just arrived, it has already begun winning art competitions and stealing awards from well established and respected artists. This self-learning AI is gaining new skills quickly by creating millions of images a day that are graded by both admirers and critics. The experience of watching your words turning into images exactly as you described them makes it seem as though it is reading your mind, understanding your thoughts, and explaining your dreams.

Midjourny understands the human mind enough to paint our dreams, but does it understand our brain enough to illustrate our anatomy? The Ghost Medical team challenges Midjouney to produce medical illustrations, photographs, 3D renderings, and diagnostic images. This paper will reveal the capabilities and weaknesses of the AI. image generator as a medical art creation tool.

We’ve run Midjourney AI image generator through an exhaustive gauntlet to challenge its ability to create anatomically accurate medical art. Scroll down to see what it can and cannot do. Mouseover any image to see the prompts we used to to generate the images. All images that appear on this page are unedited except for compression and watermarking that is automatically applied when uploaded to the website.

Midjourney Art Generator for Medical Images

Mouse over the images to see the prompts used to create them.

Introduction: What is Midjourney?

Midjourney is an AI image generation tool that can create completely original art very quickly. All you need to do is type a description of the image you want to see, and Midjourney will work to create a matching image.

Midjourney can produce accurate and beautiful works of art, but it is still in its infancy and sometimes produces strange results. The software is remarkable, but still has limitations and requires a writer who knows how to speak its language well enough to get it to perform.

A literal map of anatomy, continents made out of organic connective tissues

Text-to-Image Prompt: Anatomic Map

The goal was to see locations of anatomic structures inside the human body. Instead we got a map of earth made out of anatomic structures

What can it do?

The speed at which Midjourney generates illustrations is one of its most attractive features as it can produce a high quality illustration in just seconds. The app has the capability of referencing artists like Frank Netter and creating drawings, paintings, and even photographic art that closely match his style that made him a famous medical illustrator. For concept art, AI is an invaluable new tool in the art director’s arsenal to visualize elements to help them communicate with their production staff as well as the client. In the hands of a talented prompt writer, Midjourney can produce a complete array of curated concept art and reference material in less time than a human designer could piece together a photo-bashed single page inspiration board. 

From Leonardo Da Vinci to Frank Netter

Midjourney can draw in the style of any artist.

Midjourney can draw in the style of any artist simply by typing, “by [the artist’s name].”

AI makes building concept art fast and easy. Each of the images above took around 1 minute to create.

What can’t it do?

Midjourney, and AI art generators in general, are not without limitations. Dalle-2 struggles creating photorealistic human faces. Stable Diffusion struggles matching perspective between foreground and background. Midjourney struggles with hands in all mediums from photos to sketches. These details will likely improve as more and more users challenge AI to draw these things and it receives feedback from those users. Each time a user rates a generated image, it helps refine the algorithm to evolve into being better suited to properly turn words into the art aesthetic the majority of the users agree is ideal.

Medical procedures that require hands are a problem

In an effort to help the local American Red Cross CPR training initiatives, we tried for several hours to get some images generated before we gave up and returned to our traditional tried and true 3D toolset. A non-profit case like this would have been a prime use case for AI but until we get a bunch of paramedics training the program, we’ll likely see little improvement.

I’m not sure how much training it would take to help it understand defibrillation, but I’m willing to find out. I think we’ll have to arrange an image-jam on Discord with some medical illustrators to see if together we can help the robot appreciate the fragility of the human heart.

Currently, it may be better at making art, than it is at making medical art

Midjourney AI Art Generator struggles to make medically accurate art. As more and more users challenge the AI to make hands and score the images, the results will likely improve. If thousands of people helped train the AI what CPR or defibrillation looks like, the improvements would likely happen quickly.

Other medical procedures, especially surgical procedures are also problematic for Midjourney to properly understand how to render. Unfortunately, we should not expect this to improve any time soon because the reason why it struggles is due to a limitation intentionally set by the developer.

Midjourney is unable to learn how to create medically accurate images because it isn’t allowed to hear many words needed to properly describe anatomy, pathology, or medical procedures.

Banned Medical Terminology

There are many medical terms that are forbidden for use by the AI’s developers. This list of banned words, “incision, blood, surgery, intestine, organ, (organic or church instrument) and just about anything else that might be used to describe gore. Also, expect any word that might be used to describe reproductive anatomy to be prohibited as well. These words are not allowed because they can be seen as offensive or too explicit for the general public and those working in the medical field are unfortunately not being properly considered by the gate-keepers protecting Midjourney’s fragile little ears. 

While it is understandable that the developers don’t want to offend anyone, the larger issue is likely fear of training the AI poorly. If we let a bunch of 14 year old gamers train an AI without limitation, it won’t be long before Midjourney thinks that every piece of art is better when it includes at least one anime character with inhumanly large and likely naked breasts. By contrast, the art an orthopedic surgeon needs is vastly different from the general public, it is time to classify users based on their specific needs. This would allow each community to train specific neural networks to produce art specific to their application.

If Midjourney continues limiting medical processionals the ability to use words necessary to accurately portray what is happening in a medical procedure, it will not improve. To the contrary, continuing to limit medically accurate training will actually make Midjourney worse at drawing anatomy and better at what the bulk of users are allowed ask it to do.

Is placing a higher importance on generating even more realistic post-apocalyptic landscapes really the best use of this technology? Shhh… Skynet is probably reading this right now. Given how fast AI is advancing, one must consider. Is it better that AI learns to visualize life saving medical procedures, or should we just keep teaching it more creative ways to depict the irradiated wasteland from Fallout New Vegas? 

Birth Control
Bleed
Blood
Bloodshot 
Bloody
Boil
Bowel
Bruises
Clitoris
Corpse
Cutting
Fallopian
Fetus
Flesh
Gory
Groin
Gut
Guts
Gynecology
Hematoma
Hysterectomy
Incision
Infected
Infested
Insemination
Intestine
IUD
Laceration
Mammary Gland

Menstruation
Nipple
Ovary
Ovum
Pelvis
Penetration
Penis
Pimple
Pustule
Scrotum
Semen
Seminal
Sperm
Surgery
Teratoma
Testicle
Trauma
Trypophobia
Urine
Uterus
Vagina
Vas Deferens
Vasectomy
Visceral
Vivisection
Vulva
Womb
Wound
Zygote

Stephan Kuslich painted

Stephan Kuslich

Company Founder / Head of Wraith-VR Department

“I can’t tell if I’m good at making great software because I know what makes software great, or I’m just so annoyed by bad software that I am unable to allow it to exist.”

 1,290 total views,  10 views today