Creating realistic and natural-looking faces in Midjourney can be a challenge. Thanks to the “Vary (Region)” feature, I have found a way to enhance the realism of generated images significantly.

Midjourney excels at producing stunning visuals; however, it often struggles to generate lifelike human models, particularly from the neck up. The images can appear overly perfect, resembling beautiful cyborgs rather than authentic human beings. This flaw can be detrimental for fashion illustrators and photographers who seek realism in their work.

The distinction between a genuine photo and a typical Midjourney output often lies in the details and personality of the face. Enlarging images typically reveals a sharply rendered aesthetic, especially around the eyes and lips. The close-up image above may seem more detailed, but would you believe it’s derived from the one shown below?

It's clear which of these images was created using Midjourney:

How about these two?

Over the last 45 days, I experimented with various techniques to create a “perfectly realistic” fashion image that could easily be mistaken for traditional photography. A few weeks ago, I thought I had finally achieved that.

Thanks to extensive research and practical trials, I believe I’ve unlocked the code for realism. Here’s the secret formula that has helped me create images that are 95–97% realistic:

At the conclusion of my Midjourney prompts, I add the following phrase:

“shot with Canon R5, 70–200mm, cinematic studio lighting, strong shadows, unique angles and scene positions, photographed by Bert Stern, 8k UHD — s 800”

(Depending on the context, the lighting and shadows prompt may vary. This approach has consistently yielded my most realistic images during daylight.)

Bert Stern (1929–2013) was a renowned commercial photographer known for his iconic images of Marilyn Monroe. Midjourney seems to favor this aesthetic, producing images with a high realism factor.

However, the final elusive 2–4% was necessary for me to transition from a "nearly" perfect image to one that is virtually indistinguishable at high magnification. This is where the new “VARY (REGION)” feature proves invaluable. I've encountered numerous discussions highlighting how selecting the entire image and hitting return (without altering the prompt) yields more dynamic and realistic outputs. Some of my favorite Midjourney authors include:

• Geeky Animals
• Christie C.
• Henrique Centieiro
• Bee Lee

Inspired by their insights, I decided to test whether focusing the VARY feature solely on the face could compel the interpolation to hone in on detail in that specific area. To my delight, it worked with about 90% consistency.

Below are the Midjourney prompts and Photoshop techniques I used to consistently achieve more realistic results. In under 10 minutes, I could create images that closely resemble real life, often with slight imperfections.

Let’s dive in.

First, I set my prompt with the suffix:

“purple coverup::3 gorgeous 21-year-old woman walking on the beach in Miami, shot with Canon R5, 70–200mm, cinematic studio lighting, strong shadows, photographed by Bert Stern, 8k UHD — s 800 — style raw — ar 10:16”

This produced a somewhat realistic image:

While it’s decent, it still appeared a bit too perfect and flat.

Next, I employed the VARY (REGION) feature, selecting ONLY the face, and retyped my prompt:

“hyper detail on face::5 purple coverup::3 gorgeous 21-year-old woman walking on the beach in Miami, photographed by Bert Stern, 8k UHD — s 800 — style raw — ar 10:16”

Notice the factor of ::5—this seems to be the key.

Several of the four options were indeed varied:

The top right image, however, stood out with incredible detail (including facial lines, pores, and subtle pigment variations) and a slightly imperfect structure.

In just two minutes, I was ready to upscale.

Quick note: Some of these new images, while more realistic, may appear slightly distorted. A useful trick is to upscale the image you like, then select “VARY (SUBTLE)”—Midjourney will recreate it with the same detail while adjusting the aesthetics.

Can I use this technique repeatedly?

Absolutely! Not only did it work, but it also allowed me to enhance realism further in Photoshop. Here’s another example:

I started with the prompt:

“gorgeous 21-year-old woman walking on the beach in Miami, hyper detail on the fabric and face, shot with Canon R5, 70–200mm, cinematic studio lighting, strong shadows, photographed by Bert Stern, 8k UHD — s 800 — style raw — ar 10:16”

While satisfactory, I was aiming for near-perfection. So, I once again utilized the VARY (REGION) feature, selected just the face, and added to my prompt:

“hyper detail on face::5 gorgeous 21-year-old woman walking on the beach in Miami, photographed by Bert Stern, 8k UHD — s 800 — style raw — ar 10:16”

This led to a variety of intriguing results:

Once again, the Bottom Left image exhibited greater detail, including visible facial pores and a slightly imperfect shape.

I previously mentioned that the primary “realism” challenges lie in the eyes and face. Thus, I imported the upscaled image into Photoshop, applying an unsharp mask around the eyes and face, along with a touch of noise. The outcome was satisfying!

While I haven't reached perfection yet, I feel I'm getting closer. I eagerly anticipate what Midjourney 5.3 or 6 will offer.

Here’s another example using the prompt: “Photorealistic beautiful woman’s face::5” alongside the original Midjourney render (left). I particularly appreciate the Cary Grant chin Midjourney added during the VARY (REGION) interpolation.

I can now submit my artwork to my Wirestock.io account with the assurance that potential buyers may scrutinize my images to determine if they were generated through Midjourney. I will update you on any sales spikes on the platform in the coming month.