Midjourney prompt not producing the right image

Work through this in order

1. Shorten the prompt. Long prompts dilute weighting. Under about 40 words gives the model a clearer target.
2. Front-load the subject. Words earlier in the prompt carry more weight than words at the end.
3. Use `--no` instead of negative phrasing. "no text" often adds text; --no text removes it.
4. Check your parameters. A stray --stylize 900 or --weird value will override your intent.
5. Reset personalization. If --p or a personalization profile is active it reshapes every render; try the same prompt with --p 0.
6. Match the model version. V6 and V7 read natural language very differently from V5. Add --v 7 explicitly rather than relying on your default.

Repeatability

Add --seed 1234 to lock composition while you iterate on wording, then remove it once you like the result.