SKILL: Positive Prompt Structure

Version: 1.0 | Hermes Agent SD Prompt Craft (Technical)

DESCRIPTION

Mastery of token-weight hierarchy and clause ordering for AI image generation models. Different architectures read prompt sequences with different attention patterns — Flux weighs early tokens heavily [^57^], SDXL distributes more evenly but still prioritizes lead tokens, and LTX video requires discrete clause blocks for spatial coherence [^68^]. This skill provides the universal Subject → Action → Environment → Lighting → Style → Quality framework with model-specific adaptations.

TRIGGER KEYWORDS

prompt structure, prompt order, token weight, prompt hierarchy, clause ordering, subject first, Flux prompt, SDXL prompt, LTX prompt, prompt architecture, prompt framework, prompt template, how to structure prompt

CORE RULES

• Universal structure: Subject → Action → Environment → Lighting → Style → Quality
• Flux: first 10–15 tokens carry maximum weight — subject and key attribute go first
• SDXL: (keyword:weight) syntax available; use 1.1–1.3 range, avoid extremes
• LTX video: use discrete labeled clauses: "Subject:", "Action:", "Environment:", "Guardrails:"
• Most important attribute goes first — whatever you most need the model to get right
• Comma separation between concepts; period/line break for larger clause boundaries
• Quality tokens go last — they're important but shouldn't override subject/style position