top of page
Search

Get Consistent Characters and Styles in Midjourney V6: A Guide

midjourney blog post image
A Midjourney generated image using Midjourney Automation Suite

Midjourney keeps getting better with new features popping up regularly. Recently, there have been big updates that help you get more control over the style and even the characters in your images. This post covers the latest changes, including new ways to use style references, an improved image-to-text tool, and the exciting new character reference feature.

Exploring Midjourney's Updated Style Reference

Midjourney now has four versions of its style reference feature, known as `--sref`. This feature lets you use an image to guide the artistic look of your new creations. It copies things like colors, textures, and the overall feel from the reference image.

Here are the four versions:

  • Version 1: The first release.

  • Version 2: Became the default for a short time.

  • Version 3: An upgrade to Version 1 with better results.

  • Version 4: An upgrade to Version 2, currently the default.

You can pick which version you want by adding `--sv` followed by the version number (e.g., `--sv 4`) to your prompt. If you don't add this parameter, Midjourney uses Version 4 by default. Midjourney mentioned that this versioning might simplify in future updates, maybe with V7.

These updates make `--sref` work better. Previously, the tool sometimes pulled objects or people from the reference image instead of just the style. The new versions do this less often. Also, the `--sw` parameter, which controls how strong the style effect is, feels a bit more predictable at lower values than before.

Many users find Version 4, the default, to be quite accurate to the reference image. However, Version 3 can sometimes create more unique or surprising results, even if they don't match the reference style exactly. Experimenting with different versions can lead to interesting discoveries.

Trying out these different style reference options can take time. Consider using a tool like the TitanXT Midjourney Automation Suite to streamline testing prompts with various `--sref` parameters and versions.

Image-to-Text with Describe

Did you know Midjourney can describe images for you? The `/describe` command is a helpful tool when you are stuck trying to find the right words for a prompt. It can also give you prompt ideas.

To use it, type `/describe` followed by uploading an image or pasting an image link. Midjourney will then give you four different text descriptions of the image. If the descriptions mention artists, you can often click on their names to learn more.

You can use any of these descriptions as a prompt just by clicking a button. There's also an option to run all four at once. If you don't like the results, you can reroll for new descriptions.

These descriptions are like starting points. They won't perfectly remake your input image, but they show you words and phrases Midjourney uses to interpret it. Words that appear in multiple descriptions are likely strong keywords. This feature is great for getting ideas from images you like or understanding how Midjourney sees different elements. You can also use the `/shorten` command on `/describe` results or any prompt to see which words or phrases are most important.

Introducing Character Reference (--cref)

This is a major update! Character Reference, using the parameter `--cref`, is similar to style reference but focuses on keeping a specific character consistent across different images. This is very useful if you want to create a series of images for a story featureing the same character in different looks or situations.

To use it, add `--cref` and an image URL of the character to your prompt. Midjourney will then try to make the character in your new images look like the one in your reference image.

It’s important to know that `--cref` won't copy every small detail like freckles or shirt patterns. It aims to capture the main visual traits of the character. It works best with characters originally made in Midjourney and those with clear, distinct features. You can use it for human and non-human characters.

A cool way to use it is to put a character into different art styles, like turning a photograph into a sketch or pop art while keeping the character the same. As long as you provide a character reference, your results should include that character, even if you don't describe them much in your text prompt.

Understanding Character Weight (--cw)

You can control how strong the character reference is using the `--cw` parameter. This works a bit differently than other weight parameters.

  • The range is from 0 to 100.

  • The default is 100.

  • With `--cw 100` (the default), Midjourney focuses on matching the character's face and clothing from the reference.

  • With `--cw 0`, Midjourney only focuses on matching the character's face. This lets you change their clothes while keeping their face the same.

The key point is that Midjourney will always try to match the face from the reference image, regardless of the `--cw` value. So, use `--cw` mostly when you want to control if the clothing is also matched.

Working with --cref: Early Notes

Character Reference is a new feature, and like any first version, it has some quirks. Sometimes, you might get strange results, like unnatural body parts or poses that don't look right. Often, you can fix this by changing your text prompt. For example, rewording a prompt from "walking through" to "standing" or adding details like "looking at the camera" can help Midjourney create better images.

You might also find that even with `--cw 100`, the clothing style from the reference might not carry over perfectly, though the color might. Adding clothing details to your text prompt can help here too.

This feature will likely get better over time as Midjourney develops it. Experimenting and learning how to use text prompts alongside `--cref` is key right now.

Managing tests with different `--cref` images, `--cw` values, and text prompts can become complicated. Tools designed to automate Midjourney prompting and testing, like the TitanXT Midjourney Automation Suite, can significantly simplify finding the settings that work best for your characters.

Conclusion

Midjourney continues to add powerful tools. The new versions of style reference give you more options for artistic control, the updated `/describe` helps when you need prompt ideas, and the character reference feature is a big step towards creating consistent characters across multiple images. Keeping up with these changes helps you get the most out of the platform. Keep experimenting and see what you can create!

 
 
 

Comments


bottom of page
Midjourney Automation Suite - Automate your image generation workflows on Midjourney | Product Hunt