* This blog post is a summary of this video.

Strategies for Creating Better Faces with Midjourney AI Art Generator

Author: Scott DetweilerTime: 2023-12-30 16:35:01

Table of Contents

Using Redundant Descriptors to Improve Face Rendering

One key strategy for getting better faces out of AI image generation is using redundant descriptors. This means using multiple descriptive terms related to the face, such as calling it beautiful, lovely, cute, slight smile, etc. The more terms you use to describe the type of face you want, the more the AI renderer has to focus on getting those facial qualities right.

For example, instead of just saying you want a picture of a medusa, say you want a picture of a "beautiful medusa with a slight smile and lovely face". This forces the AI to ponder what you really want in the facial rendering and make that a priority.

Beautiful, Lovely, Slight Smile

Using descriptive terms like beautiful, lovely, cute, and slight smile are ways to call specific attention to the kind of facial qualities you want the AI rendering to prioritize. The AI will see those terms and be prompted to spend more time getting the facial qualities to match what was described.

Symmetrical, Single Face

Other helpful descriptors for faces are symmetrical and single face. These terms help prompt the AI to render just one face that is properly aligned and symmetrical, rather than accidentally generating extra or misaligned facial features.

Providing Example Images to Guide the AI

Another way to improve facial rendering is providing example images of the kind of face you want. For instance, name a specific actress and say you want a face like hers. This gives the AI actual facial examples to train on.

When you provide a specific person for facial reference, the AI will look at many pictures of that person online and try to emulate attributes like their eye shape, nose shape, lip shape, etc. Just make sure to choose someone who is very widely photographed on the internet for best results.

Iterating and Re-rolling for Better Results

It is rare to get exactly what you want from an AI image generation prompt on the very first try. Be prepared to re-roll and iterate on the prompt over and over until you get closer to the desired outcome.

Don't be afraid to keep the base prompt similar but change small things in each iteration, like trying a different example face or adding/removing some descriptive facial terms. Over many rolls of the dice, you will inch closer to what you really wanted to create.

Upscaling for Increased Detail and Resolution

Most AI image generation starts off quite small in resolution and detail. But services like Midjourney include upscaling capabilities to take a small AI rendered image and enlarge it while algorithmically enhancing quality and details.

Be sure to upscale any smaller AI images to get crisper facial details. Then you can see where flaws may still exist, versus just assuming the face came out poorly due to low resolution.

Enhancing Further with Tools Like TenCent Arc

If you want to improve AI rendered faces, but don't have advanced photoshop skills, look into automated tools like TenCent Arc. This is a Chinese web application that specifically fixes problems with faces in images.

You simply upload your AI image, TenCent Arc runs an algorithm to repair any oddities or asymmetries in the face, then sends you back a cleaned up version. This can be great for getting naturally pleasing faces out of otherwise flawed AI renderings.

Turning AI Art into Unique Creations

Remember that AI image generation platforms create art, but don't necessarily make it your unique creation alone. To make it your own, look for ways to build on the AI art with custom changes.

Some ideas are combining it with your own photography in Photoshop, printing it out and adding real paint or touches by hand, bringing it into digital painting software to alter further, and more. Get creative with making the base AI art your own!


Q: How can I get the AI to focus more on rendering a good face?
A: Use redundant descriptors like "beautiful", "lovely", "cute" etc. and details like "slight smile" or "symmetrical face".

Q: What if I want the face to resemble a certain person?
A: Provide an example image of that person in the prompt to guide the AI, but don't use copyrighted images without permission.

Q: What should I do if I don't like the initial results?
A: Click the "re-roll" button to have the AI generate different variations. Keep iterating until you get something promising.