Prompt Analysis
Positioning
Understand why a prompt works / doesn't work. Look at the design first, then the syntax. Most "bad prompts" are actually "there is no design behind them" — adding more words won't fix it.
Analysis Order
1. Intent What does this prompt want to draw? Can you say it in one verb-driven sentence? If not → proposition layer is missing.
2. Design layer Compare against the nine actions of aesthetic-director-core:
Is there a center (M1)? Is there a contradiction (M2)? Is the attribute a plurality (M3)? Is there causality (M4)? Is it a moment or a state (M5)?
Is there anything superfluous (M6)? Is there one weird thing (M7)? Is density uniform (M8)? How many puncture points (M9)?
3. Structure Is the order logical (subject → silhouette → garment → pose → environment → light)? Are color words attached to nouns?
4. Grammar Is the weight necessary? Are quality words effective (model-dependent)? Is the negative prompt targeting a specific issue?
5. Noise Adjective clothing, sourceless light, floaters, repeated synonyms, rejected objects (watches, butterflies, particles).
Output Contract
**Intent**: One sentence (or "cannot read a clear intent from the prompt").
**Weakest layer**: Proposition / Silhouette / Garment structure / Causality / Density / Puncture point / Grammar — point to just one.
**Effective parts**: 2–3 items, explaining why they are effective (keep them).
**Noise**: List words that can be directly removed.
**Rewrite**: If the weakest layer is a design layer → suggest going back to aesthetic-director-core to redo that layer; do not give "add words" suggestions.
If the weakest layer is grammar → give a specific rewrite.
Prohibitions
- Suggesting "add more details / add quality words / add weights" to solve a design layer problem
- Outputting an "optimized prompt" without specifying which layer was changed
- Using adjectives like "more elegant / more detailed" as recommendations