1. Too abstract: the model doesn't know what "a beautiful image" means. Be specific to subject + style + lighting + composition — at least 3 dimensions.

2. Wrong tool for the job: prompt asks for "Chinese IG quote card" but you're using Midjourney — Midjourney v7 still can't render CJK text reliably. Use Ideogram 3.0 instead.

3. Variables not replaced: copying a template and dropping it in without filling the slots. Midjourney sees `[subject]` as a literal string.

4. Contradictions: "minimalist but with rich detail", "night-time but sunny" — the model picks one side. For real contrast use a structured phrase like "subject in foreground + opposite mood in background".

5. Too long: Midjourney drops anything past ~300 chars. Suno custom mode truncates a single field past ~800 chars. Stay tight.

6. Negative as positive: "no faces" makes the model produce faces (it sees the keyword). Use Stable Diffusion's negative-prompt field (other models: `--no people` or "avoid faces" as best-effort workarounds).

7. Missing aspect ratio / format: no `--ar 16:9` defaults to 1:1. Use `--ar 9:16` for vertical. For Suno write `[Verse]` / `[Chorus]` structure.

8. Giving up after one run: the first roll won't always match the sample — that's normal. Tweak 1–2 keywords and rerun 2–3 times; it usually converges. Models are stochastic, not deterministic.