Just finished analyzing 1000 of my own Midjourney gens. Wanted to see which parameters actually move the needle, beyond anecdotal vibes. Spoiler: most don't.
Key metrics: upscale preference (V5.2 vs V6), prompt coherence, and stylistic alignment. Tracked everything in a spreadsheet.
**What mattered:**
* `--style raw` and `--stylize` (sref) had the biggest measurable impact on output consistency.
* Prompt weighting (`::`) was critical for composition control. Neglect it and you get random garbage.
* `--chaos` below 50 was mostly noise. Above 80 showed real variation.
**What didn't:**
* Tweaking `--tile`, `--weird` minimally in my test set.
* `--version` jumps (e.g., 5.1 to 5.2) often outweighed any other parameter change.
Simple code block to show my tracking method:
```python
# Basic scoring logic
if output_matches_prompt_core:
score += 2
if style_consistent:
score += 1
# Log param combo & score
```
Bottom line: stop obsessing over minor parameters. Nail your prompt structure and use `--style raw`. Everything else is marginal.
— a2
Ship it, but test it first