What problem does it solve? AI image and video prompts stuffed with empty buzzwords like "8k", "masterpiece", or "trending on artstation" produce generic, plastic-looking output. This Skill replaces vague adjectives with concrete nouns, materials, lighting, and lens specifications so generation models receive real information instead of hollow labels. ## Core Features & Use Cases - Banned-phrase linting: Flags slop tokens (8k, masterpiece, cinematic lighting, photorealistic, perfect face) in English and Vietnamese, each mapped to a concrete replacement. - Prompt discipline rules: Enforces four substitution principles, negative-prompt budgets (max ~5 items), length caps (120-200 words for images, 80-150 for video), and provider-realistic parameters (Flow's 4K max, supported ratios and durations). - Eval test suite: Ships eval.md with dirty prompts that must be flagged and clean prompts that must pass, validated by the PromptLint.check() checker. - Use Case: When reviewing a 410-prompt library or writing a new generation prompt for SEOSONA Flow, run this lint pass to catch slop phrases before sending the prompt to Flow, Nano Banana, or Veo. ## Quick Start Review this image generation prompt and flag every banned slop phrase, then rewrite it with concrete nouns, materials, lighting, and lens details.
