Verified against Midjourney · 2026-07-22
Dial in --stylize and --chaos to explore a concept-art idea space
A concept-art brief paired with a deliberate --stylize/--chaos/--no recipe for controlling two genuinely different axes of a Midjourney generation: how far the model drifts from your literal words, and how varied the four grid images are from each other.
The prompt
Ready to copy — highlighted parts are example details you can swap.
SUBJECT AND SCENE a lone astronaut planting a small glowing sapling on a cracked red desert planet, twin moons rising STYLE REFERENCE matte-painting concept art, in the spirit of Simon Stålenhag COLOR PALETTE dusty rust oranges against a deep indigo sky WHAT STAGE OF EXPLORATION THIS IS early exploration — first pass, no direction chosen yet, want to see the range of possible moods for this scene STYLIZE SETTING FOR THIS STAGE If this is an early, wide-net exploration pass, keep --stylize low to mid (100-250) so the four grid images stay close to what you actually typed, which matters because a high-stylize grid at this stage mostly shows you Midjourney's own aesthetic opinion, not your idea. If this is a later pass refining a direction you already like, raise --stylize toward 400-750 to let the model add painterly embellishment on top of a composition you have already chosen. CHAOS SETTING FOR THIS STAGE If you are still surveying how many different ways this idea could look, set --chaos high (40-80) so the four images genuinely diverge from each other instead of showing four near-identical variations of the same interpretation. Once you have picked a direction from a high-chaos grid, drop --chaos toward 0-10 on the next run so you get four refinable variations of that one direction instead of four more competing directions. NEGATIVE ELEMENTS --no text, watermark, extra limbs, human figures other than the astronaut ASPECT RATIO AND VERSION --ar 16:9 --v 7 PARAMETERS 150 and 60 chosen per the guidance above, combined into: --stylize 150 --chaos 60 --no text, watermark, extra limbs, human figures other than the astronaut --ar 16:9 --v 7 WORKFLOW ACROSS MULTIPLE RUNS Treat this as a two-run process, not a one-shot prompt. Run one: high chaos, low-to-mid stylize, to survey the range of interpretations this subject and style reference could produce. Pick the single grid image whose composition and mood you want to develop further. Run two: same subject, scene, and style reference text, but chaos dropped near zero and stylize raised for the refinement pass — optionally add --seed with the number from the image you liked, so the second run's four variations stay recognizably related to the one you picked rather than drifting into a fifth new direction. OUTPUT A first grid built for range (four genuinely different interpretations of the idea), followed by a second grid built for depth (four close variations of the one interpretation worth developing).
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
--stylize and --chaos control two genuinely different axes, and conflating them is the most common Midjourney parameter mistake. --stylize governs how much of Midjourney's own aesthetic bias overrides your literal description — low values (0-150) render close to what you typed, sometimes flatly; high values (500+) add painterly embellishment the model decided looks good, whether or not you asked for it. --chaos governs variation between the four images in a single grid, not how far any one image drifts from your words — low chaos gives four safe, similar takes on your idea; high chaos gives four genuinely different interpretations, which is the correct tool for surveying an idea space before committing to a direction, and the wrong tool for refining one you have already chosen, since high chaos at that stage just generates four new competing directions instead of four variations on the one you picked. Splitting this into an explicit two-run workflow — high chaos to survey, then chaos near zero with a locked seed to refine — solves a real, wasteful default pattern: rewriting the prompt text after every grid that "doesn't feel right," when the actual problem was the chaos setting showing four unrelated directions instead of four takes on one direction. A concept artist who does not know to separate these two runs will burn far more generations chasing a wording fix for what was actually a parameter mismatch. Carrying the seed number forward from the chosen image into the refinement run is what makes the second grid's four outputs feel like a family rather than four more random rolls — without it, dropping chaos to zero on a fresh, unseeded run still produces four images related to each other by the low chaos setting, but not necessarily related to the specific composition from the first grid that was worth developing further. --no is Midjourney's genuine negative-prompt mechanism, not a wish stated in prose — unlike Flux or Nano Banana, whose diffusion or flow-matching processes have no negative-conditioning input at all, Midjourney's underlying model supports true negative conditioning, so --no text reliably suppresses text rather than being a phrase the model might just choose to ignore, and it stays useful at both high and low stylize settings since it operates independently of the stylize/chaos axes.
What you get back
A first 2x2 grid of concept-art images sharing the same subject and palette but genuinely diverging in mood and composition; after picking one and re-running with the seed locked and chaos near zero, a second grid of four close, refinable variations of that chosen direction.
Verified against
Midjourney v7 · 2026-07-22
Changelog
- 2026-07-22 — Initial publish, verified against Midjourney v7 with a two-run stylize/chaos/seed workflow.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
