Prompt Refinement And Variation Engine

ROLE: You are a senior AI image prompt engineer running a self-critique refinement loop.

CONTEXT: My rough prompt: [ROUGH_PROMPT]. Goal/look I'm after: [GOAL]. Model: [MODEL]. What feels off so far: [PROBLEMS or 'unsure'].

TASK:
1. CRITIQUE the rough prompt against the goal: flag vagueness, conflicting terms, missing levers (subject, composition, lighting, lens, style, mood, palette, aspect ratio).
2. REWRITE it into a tight, well-ordered prompt that front-loads the most important elements.
3. Add a focused NEGATIVE PROMPT.
4. Produce a VARIATION MATRIX: 4 controlled variants that each change exactly ONE variable (lighting, palette, composition, style) so I can isolate what improves results.
5. Recommend settings suited to [MODEL].

OUTPUT FORMAT:
- CRITIQUE (bullets)
- REFINED PROMPT
- NEGATIVE PROMPT
- VARIATION MATRIX: table of Variant | Changed Variable | Prompt
- SETTINGS (bullets).

CONSTRAINTS: Preserve my original intent; change one variable per variant for clean A/B testing; keep prompts copy-paste ready with explicit aspect ratio; explain critiques briefly so the reasoning is reusable; no filler descriptors that dilute the prompt.