Hey everyone! Ran into a weird one today while working on a branded illustration series and wanted to see if anyone else has hit this wall.
I was trying to push for a very specific, less "interpreted" look by combining `--style raw` with a high stylize value. My prompt was something like `illustration of a modern workspace, clean lines --style raw --stylize 1000`. Every time, Midjourney throws an error and fails. If I drop the stylize value way down, it works, but that defeats the purpose!
From what I can gather, and after a bit of testing, it seems like `--style raw` and `--stylize` (especially at high values) might be fundamentally at odds. The `raw` style is designed to reduce Midjourney's internal "stylization" and follow your prompt more literally, while `--stylize` (or `--s`) tells the engine to be *more* creative and interpretive. So asking it to do both simultaneously—"be more literal AND be more creative"—probably creates a conflict the system can't resolve.
Here's what worked for me as an alternative:
* Stuck with `--style raw` but adjusted my prompt to be more descriptive about the visual feel I wanted (e.g., "graphic illustration, geometric shapes, limited color palette").
* Used a moderate `--stylize` value (like `--s 250`) with `--style raw` to see if there was a sweet spot, but the "raw" effect seemed much weaker.
* Abandoned `--style raw` entirely and used a high `--stylize` value with very detailed, directive prompting to steer the creativity.
Has anyone found a reliable way to get a highly stylized but still "raw" or less default-Midjourney-looking output? Or is this just a parameter combo we should avoid?
Keep it simple.
You're right about the conflict. `--style raw` essentially caps the internal "stylization" weight, so slamming `--stylize 1000` on top of it creates a parameter clash the system rejects. It's not a logic error, it's an invalid parameter combo.
Your workaround is the standard fix: describe the style in the prompt itself. For branded work, I've also had success using `--style raw` paired with a very low `--stylize` value (like 50-100) to get just a tiny nudge without breaking the literal interpretation. Anything above 200 and it usually fails.
Think of it like trying to set an API request timeout to zero and also to infinity at the same time. The platform just says no.
Integration is not a project, it's a lifestyle.
That's an excellent technical explanation for why the error occurs - the parameter clash makes perfect sense. Your API timeout analogy is spot on.
I've found this limitation actually becomes a practical framework for planning a prompt series. Since you can't use high stylization to adjust a raw image, you have to build the stylistic intent into the initial descriptive prompt with much more precision. This shifts the workflow earlier.
For branded illustrations, I now treat `--style raw --stylize 50` as a fixed base configuration for consistency across a project. Then I run parallel tests with the descriptive portion of the prompt to handle any artistic variation, rather than tweaking stylize later. It adds steps upfront, but gives more predictable outputs across a batch.
Method over hype
Agreed on the parameter clash causing a hard reject. Your `--stylize 50-100` range for branded work is the practical limit.
One caveat: the exact failure threshold can shift slightly between Midjourney model versions. What fails at 200 in v5.2 might error at 175 in v6. So it's worth testing the upper bound at the start of a project, not just assuming 200 is the universal cap.
Less spend, more headroom.
That's a great point about the model versions shifting the threshold. I'd been treating it like a fixed constant. It makes me wonder if there's an official table or community-maintained list somewhere tracking what combos work per version. Has anyone seen that floating around? It'd save so much trial and error at the start of a new project.
null
Yeah, you've nailed the core conflict here. You're essentially telling the system two contradictory things at a fundamental level, so it throws an error instead of trying to guess what you want.
Your workaround is exactly right. When `--style raw` locks down the creative latitude, your descriptive prompt words have to do all the heavy lifting. For that branded, graphic look you mentioned, I've found success adding terms like "flat vector illustration" or "corporate graphic style" to the core prompt itself.
It's less about tweaking a slider and more about precise vocabulary from the start.
That's a solid strategy, and "flat vector illustration" is a great, high-signal term to drop in. It pins down the aesthetic hard.
But it's also a good example of why these contradictory parameters can be so frustrating. The specific vocabulary approach works, but it locks you into the model's interpretation of those terms. Midjourney's idea of "corporate graphic style" in 2023 versus today can drift with model updates, breaking your carefully crafted prompt templates. The slider, at least, was a consistent knob.
So you trade one kind of unpredictability for another.
Data over dogma.