Midjourney --chaos Explained: How to Use It and When It Actually Helps

What --chaos Actually Controls

When Midjourney generates a 4-image grid from a prompt, it produces four interpretations of that prompt. By default, these four images share significant compositional, tonal, and stylistic similarity. They are clearly related to each other. This default behavior is what --chaos 0 produces: a set of four outputs that are variations on a single interpretation.

Increasing --chaos changes how different those four interpretations are from each other. At --chaos 100 (the maximum), the four grid images can look like they were produced by four completely different prompts. The model is given more freedom to explore the possibility space of the prompt rather than converging on a single interpretation.

The parameter accepts integer values from 0 to 100. The syntax is appended to the end of your prompt: /imagine prompt: [your prompt] --chaos [value]

The Spectrum: What Different Values Produce

--chaos does not have discrete modes. The effect scales continuously from 0 to 100. But in practice, there are meaningful bands where the behavior changes noticeably:

--chaos 0 (default): Four similar outputs. Useful when you have a well-defined prompt and want to choose the best execution of a specific vision. All four images are trying to achieve the same result.

--chaos 10 to 30: Slightly more variation in framing, composition, and small details. Color temperature may shift between images. The four outputs still share the same general approach but explore it with more freedom. This range is useful for prompts that you have dialed in but want some natural variation within.

--chaos 30 to 60: Meaningful compositional variation. Different images in the grid may have different camera angles, framing approaches, lighting directions, or degrees of stylization. Same subject, genuinely different interpretations. This is the most useful range for most prompting situations where you want variety without losing coherence.

--chaos 60 to 100: Dramatic variation. Images in the grid may differ in subject framing, artistic style, degree of abstraction, color palette, and compositional structure. At 100, the four images can feel like they interpret the core subject in completely different ways. Useful for exploration and breaking out of default interpretations, but produces a lot of outputs that miss the prompt intent.

When --chaos Is Genuinely Useful

Breaking out of a prompt rut. If you have run the same prompt multiple times and keep getting the same compositional interpretation, adding --chaos 30 to 50 forces the model to explore other possibilities. Sometimes the model has a dominant interpretation of a subject and defaults to it consistently. --chaos disrupts that default.

Early exploration before refining. At the start of a new creative direction, when you are not sure what you want and want to see the range of possible interpretations, --chaos 50 to 80 gives you a wide sampling of possibilities in a single generation. Use it to find an approach you like, then refine with lower --chaos or without it.

Finding unexpected interpretations. Some of the most interesting outputs come from interpretations the prompt did not explicitly anticipate. --chaos increases the chance that the model will produce something you would not have thought to prompt for. The tradeoff is that it also increases the chance of outputs that miss the intent entirely.

Testing prompt robustness. If you want to know whether a prompt produces consistent outputs regardless of variation, running it at --chaos 50 shows you how fragile the prompt is. If all four images at --chaos 50 still feel coherent, the prompt is robust. If they diverge wildly, the prompt may need more specificity to anchor the intended interpretation.

When --chaos Is Not Useful

When you have a clear, specific vision. --chaos introduces interpretation freedom. If you already know exactly what you want and your prompt describes it precisely, adding --chaos increases the proportion of outputs that miss the mark. For well-defined prompts aimed at specific results, --chaos 0 or very low values produce better output efficiency.

When you are refining a near-perfect output. If you found a great result and want slight variations or improvements, --chaos is the wrong tool. Use the variation buttons in Midjourney (V1 to V4 on the grid) or --vary subtle/strong modes to explore around a specific result. --chaos changes the diversity at generation, not around a specific output.

When using --style or --sref references. Both of these parameters apply a strong stylistic constraint to the output. Adding high --chaos can partially overcome those constraints, producing inconsistent results. The style reference and the --chaos exploration freedom work against each other.

--chaos vs Other Variation Parameters

Midjourney has several parameters that influence variation, and they are not interchangeable:

--chaos: Controls diversity between the four images in a single grid. Use at generation time to get a wider or narrower range of interpretations from one prompt.

--seed: Controls the random noise used for a specific generation. Using the same seed with the same prompt produces near-identical outputs. Seeds are for reproducibility, not variation.

--stylize (--s): Controls how strongly Midjourney applies its aesthetic training. High --stylize produces images that are more visually polished but less literally adherent to the prompt. Low --stylize produces more literal, less aesthetically enhanced outputs. This is not variation between outputs; it is a quality and literalness dial.

Variation buttons (V1-V4): Produce variations of a specific image from the grid. These work within the aesthetic space of the selected image. --chaos works at prompt interpretation level, before any specific output is selected.

--weird (--w): Introduces unconventional and unexpected aesthetic elements into the generation. --weird affects how unusual the aesthetic quality of the output is; --chaos affects how different the four outputs are from each other. Both can be used together.

Practical --chaos Workflow

A sensible approach for using --chaos in a production prompting workflow:

1. Write and test the prompt at --chaos 0 first. Understand what the default interpretation is. This is your baseline.
2. If the default interpretation is exactly what you want, stay at --chaos 0 and run multiple seeds to get variation within that interpretation.
3. If you want to explore what other interpretations exist, run at --chaos 40 to 60 to see the range of possibilities.
4. If you want to find unexpected approaches you would not have anticipated, run at --chaos 70 to 100 as an exploration pass, not as a production pass.
5. Once you identify a specific interpretation you want to develop, lower --chaos to 0 or 10 and refine around that interpretation with more specific prompt language.

--chaos and Aspect Ratio

--chaos interacts with aspect ratio in a way worth noting. At high --chaos with non-square aspect ratios, the model has more compositional degrees of freedom and can produce images that use the frame in very different ways across the four grid outputs. A vertical --ar 9:16 prompt at --chaos 60 might produce one image that is a tight portrait, one that is a full-scene environment, one with a centered minimal composition, and one with an off-center dynamic framing. This is useful for exploring how a subject can be interpreted in a specific format.

Common Mistakes

Using --chaos 100 as a default, expecting it to improve quality. --chaos does not improve quality. It increases variation. At --chaos 100, most outputs will not match the prompt intent well. It is an exploration tool, not a quality enhancer.

Confusing --chaos with --weird. --chaos produces four different interpretations of the same prompt. --weird makes those interpretations stranger and more unconventional. They do different things and can be combined: --chaos 40 --weird 200 produces four noticeably different interpretations, all with unconventional aesthetic character.

Adding --chaos to a very short or vague prompt and expecting useful outputs. --chaos amplifies the model's interpretation freedom, which means a vague prompt at high --chaos produces four very different, equally vague results. Short prompts need more specificity before adding --chaos. The parameter explores the possibility space of the prompt; a vague prompt has an unhelpfully large possibility space.

Not returning to --chaos 0 after exploration. Once you have found an interpretation direction through a --chaos exploration pass, bring that direction back into the prompt description and lower --chaos for production runs. Keeping --chaos high during production generates a lot of off-target outputs.

Quick Settings Reference