Midjourney --chaos Explained

Defining Chaos

“We live in a rainbow of chaos.”

— Paul Cézanne

First, let's figure out what --chaos is and how it works in real life. Here is the official definition from Midjourney Documentation:

The --chaos or --c parameter influences how varied the initial image grids are. High --chaos values will produce more unusual and unexpected results and compositions. Lower --chaos values have more reliable, repeatable results.

The default --chaos value is 0, and you can go as high as 100. Sounds good⁠—let's test it!

In short: --chaos influences a generation in relation to the other three variants. Metaphorically speaking, --chaos 0 is a launchpad where four average/base Midjourney variants assemble in response to your prompt. When you increase --chaos, it launches those variants away from each other.

The higher values will generate “crazier” results⁠—because Midjourney is forced to generate the outcome farther and farther away from the “average center.”

Controlling Chaos

An interesting thing happens to --chaos when you lock the --seed parameter.

Midjourney will still try to push the variants away from each other, only now it has to do it within stricter limits.

And the first thing you notice⁠—in locked --seed situations, --chaos seems to actually add order to the variations!

Someone with a stronger will, more dedication, time, and a technical mindset could squeeze better results out of some of the methods in this study.

Stylizing Chaos

But isn't there already a way to play around with your generations within a locked --seed parameter? Like, --stylize? Yes! Let's see how the two parameters work together (and how stylize works alone).

--stylize influence is visible, but the changes are somwhat subtle.

Let's try increasing --chaos now!

Turns out, --chaos is no less efficient way to experiment and enrich your generations than --stylize, and they do work well together!

Applying Chaos

Here is another interesting effect of raising the --chaos value. Not only do the higher values affect the variations' subjects. On top of that, Midjourney styles tend to become more varying in the way they are applied.

In this example, higher --chaos added more variability to Tyrus Wong's base prompt and broadened its spectrum (note the difference in the painting manners in the --chaos 99 sample).

Varying Chaos

For some time already, we use low --chaos values when generating daily Benchmarks for our Instagram. The reason? Sometimes, it makes "weaker" styles show themselves!

Let's observe how, in this final example, --chaos helped reveal the hidden potential of Neri Oxman's name.

Starting from 66, there is a much more distinct Oxman-style variant that wasn't around in default generation! And look at the coolest #4 at --chaos 99!

Instead of Conclusion

I hope this study will⁠—counterintuitively ;)⁠—make your Midjourney workflow more controllable, and your MJ toolbox will expand! Did you have an interesting experience with --chaos that you want to share? Want to correct me or add something to the study? Send me a note via the form below!

Happy midjourneys!
‍
— Andrei