Midjourney — Parameters & Its Own Artistic Bias
Generative AI Prompting for Image Models
Chapter 4 · Midjourney — Parameters & Its Own Artistic Bias
imgai1-3's six-category vocabulary is shared across every tool in this course — but how that vocabulary is delivered to the model, and what the model does with it once it has it, differs a great deal by tool. This chapter is the first of three (imgai1-4–imgai1-6) covering that tool-specific layer, starting with the one running the most cloud-only, most curated, most parameter-driven of the three.
The Discord-Based Workflow
Midjourney has no standalone desktop app and no local install — it runs entirely in the cloud, and (traditionally) was accessed through Discord: joining Midjourney's own Discord server (or inviting its bot to a private server), typing /imagine followed by a prompt in a text channel, and receiving a 2×2 grid of four candidate images a short time later. A newer web-based interface now exists alongside Discord, but the underlying workflow is the same: submit a prompt, get four options, then either upscale one, generate variations of one, or rerun the whole prompt.
imgai1-5), Midjourney doesn't publish its model architecture or training details. Everything in this chapter about what a parameter "does" reflects Midjourney's own published documentation and consistent, widely observed community behavior — not an academic description of internal mechanics the way imgai1-2's diffusion explainer could be for the general case. Keep that distinction in mind: this chapter describes what Midjourney reliably does, not exactly how it does it internally.
Parameters — Appended, Not Woven In
Midjourney prompts follow a consistent shape: the descriptive prompt itself (built from imgai1-3's six categories), followed by one or more --parameter flags appended at the end. Parameters configure the generation process itself rather than describing the image's content.
a weathered lighthouse on a rocky cliff, oil painting, dramatic golden-hour lighting --ar 16:9 --stylize 250 --chaos 15
| Parameter | Controls | Notes |
|---|---|---|
| --ar (aspect ratio) | Output image proportions | e.g. --ar 16:9 for widescreen, --ar 1:1 for square (the default) |
| --stylize / --s | How strongly Midjourney's own trained aesthetic bias is applied vs. literal prompt adherence | Range roughly 0–1000; low values track the prompt more literally, high values lean into Midjourney's own house style |
| --chaos / --c | How much the four initial grid results vary from each other | Range roughly 0–100; low values produce four similar takes, high values produce four genuinely different interpretations |
| --no | Excludes specified content — Midjourney's own negative-prompt mechanism | --no text, watermark; the tool-specific implementation of the negative-prompt concept covered generally in imgai1-7 |
| --iw (image weight) | How strongly an attached reference image influences the result relative to the text prompt | Used alongside an image URL provided directly in the prompt itself |
--stylize 0 tends to produce something close to a literal, undecorated rendering of that description. The same prompt at --stylize 750 often comes back noticeably more dramatic, atmospheric, and painterly than the prompt itself asked for — extra lighting, extra mood, extra visual polish the text never requested. This is the mechanism behind this chapter's next section.
Midjourney's Own Well-Documented Artistic Bias
Midjourney is widely and consistently observed — by its own users, in its own documentation, and across independent comparisons — to lean toward a dramatic, painterly, highly polished aesthetic even when a prompt doesn't ask for one. Flat, neutral, purely documentary-style requests often come back with cinematic lighting, rich color grading, and a level of visual "finish" the prompt never specified. This isn't a bug or an inconsistency; it's a real, deliberate product characteristic, widely understood to result from how Midjourney curates and fine-tunes its own model toward outputs its own community and internal review consistently rate as more visually striking.
This matters directly for how you write Midjourney prompts: getting a genuinely neutral, unstylized result on purpose usually takes deliberate, explicit effort (a low --stylize value, explicit style-suppressing language) rather than simply omitting style descriptors and expecting a blank slate.
| Midjourney | What this sets up | |
|---|---|---|
| Default tendency | Dramatic, painterly, highly polished, even unrequested | Contrasted with Stable Diffusion's more neutral baseline (imgai1-5) and DALL-E's own different tendency (imgai1-6) |
| Access model | Cloud-only, Discord/web, no local install | Contrasted with Stable Diffusion's own local-hosting capability (imgai1-5) |
| Openness | Closed-source, documented behaviorally | Contrasted with Stable Diffusion's own fully open architecture (imgai1-5) |
Hands-On Exercises
Explain what --stylize controls, using this chapter's own tip-box example (the "red cube" prompt at --stylize 0 vs. --stylize 750), and explain why the difference between those two outputs is a real, deliberate product characteristic rather than random inconsistency.
📄 View solutionUsing this chapter's own warn-box, explain why this chapter can only describe what Midjourney's parameters do, not exactly how they work internally, and contrast this with how imgai1-2 was able to describe diffusion models in general.
📄 View solutionExplain the practical consequence of Midjourney's own artistic bias for someone trying to get a genuinely neutral, undecorated result — what do they actually have to do, and why doesn't simply omitting style descriptors work the way it might on a more neutral tool?
📄 View solutionChapter 4 Quick Reference
- Workflow: Discord (or web) —
/imaginea prompt, get a 2×2 grid, upscale or vary - --ar aspect ratio · --stylize literal-vs-house-style · --chaos variation across the initial grid · --no exclusion (tool-specific negative prompt) · --iw reference-image weight
- Midjourney has a real, documented bias toward dramatic/painterly results, even unrequested — getting neutral output takes deliberate effort
- Closed-source — this chapter describes documented behavior, not internal architecture
- Next chapter: Stable Diffusion — Open-Source Control & Technical Parameters