Easy Guide: Create Consistent Characters in Midjourney

Getting Midjourney to create the exact same character across different images can be tricky. Each time you enter a prompt, you usually get something brand new. This makes it hard if you want to tell a visual story or show a character in different situations.

People have tried many ways to solve this. This post looks at an experiment comparing Midjourney Version 4 and Version 5 to see which one handles character consistency better and what techniques might help.

Why Character Consistency Matters

Imagine you are creating a comic strip, a series of illustrations, or images for a game. You need the main characters to look like the same person in every picture. Without consistency, your project loses that connection.

Finding reliable methods means less time generating and regenerating images, trying to get a match.

streamline your Midjourney workflow, especially for tasks requiring character consistency, consider using the Midjourney Automation Suite from TitanXT. It can help manage prompts and variations more efficiently.

Testing the Waters: V4 vs. V5

The experiment started with a character based on a "Vampire Woman" prompt. Simple changes were made to the prompt, like altering hair color (from ashen to golden wavy) or hair color and eye color (ashen hair, bright green eyes). The goal was to see if Midjourney would keep the character's overall look the same despite these minor prompt tweaks.

Then, a name ('Ashley') was added to the prompt. The prompt was run in Midjourney Version 4. These first results showed some similarity, suggesting V4 might be capable of consistency.

Using Images as Seeds

A common technique is to use a generated image as a seed for future prompts. The experiment took one of the V4 images and used it as a seed image link at the start of a new prompt. This prompt asked for a side pose of the character 'Ashley.'

Running this in Version 4 produced results that were getting closer, but still not a perfect match. Trying another version of the pose prompt helped a little more. Reducing some of the extra descriptive text in the prompt also seemed to improve consistency slightly in V4.

V4 Shows Promise

The tests in Version 4 suggested it was doing a better job of picking up on the character's details and producing similar output. Even when the prompt was changed completely to 'Ashley fighting with a sword with untied hair with werewolves,' the V4 output was quite close to the original character.

Managing multiple character variations and seeds manually can become complex. Explore how the Midjourney Automation Suite from TitanXT can help organize and automate your consistent character generation process.

How Did V5 Perform?

Trying the same character and prompt variations in Midjourney Version 5 gave different results. V5 tended to produce very realistic images, which might not align with the desired style. More importantly, the characters generated in V5, even using some techniques, seemed to change much more dramatically from one prompt to the next compared to V4.

Running a new test with a different character name ('Judy') confirmed this. The initial V5 results for 'Judy' did not match each other well. Using one of the 'Judy' images as a seed in V5 also failed to produce consistent results when asking for an action scene ('Judy fighting with werewolves with a sword'). The output was often cartoonish or showed a completely different character.

The Importance of the Front-Face Shot

The most successful attempt at consistency in the experiment came when using a crisp, front-facing headshot of the character as the seed image. When the original 'Ashley' headshot (which was a clear front view) was used as the image seed for the action prompt ('Ashley running and fighting with a sword...'), the results were much closer to the original character.

This suggests that when trying to maintain consistency, having a clear front view image of your desired character is a key starting point, especially when using image seeds.

Wrapping Up Findings

Based on this experiment, Version 4 appears to be better suited for creating consistent characters than Version 5. The most effective technique found was using a clear front-face headshot image of the character as the seed image for subsequent prompts.

Consistency in AI image generation requires specific strategies. Tools designed for batch processing and workflow management can save significant time. Automate your consistent character creation using the Midjourney Automation Suite from TitanXT.

Achieving perfect consistency is still an area being explored, but using V4 with strong front-face image seeds seems like a solid method to get much closer results.