SKILL: Photometric Overexposure

Version: 1.0 | Hermes Agent Failure Pattern Library

DESCRIPTION

Mastery of preventing and correcting overexposure in AI-generated imagery. Unlike camera overexposure (where RAW data can be recovered), AI overexposure represents missing data — the model never rendered detail in blown areas [^77^]. Recovery requires regeneration with lighting-aware prompts rather than post-processing. This skill provides prevention vocabulary, detection signals, and fix strategies.

TRIGGER KEYWORDS

overexposure, blown highlights, white blowout, too bright, burned out, white out, overlit, highlight clipping, bright white area, lost detail, exposure problem, washed out, bright light problem

CORE RULES

• AI overexposure cannot be fixed in post — the detail was never generated; regenerate with fix vocabulary
• Prevention: specify "controlled exposure, highlights retained, no blown areas"
• Name the light source and reduce its intensity: "soft ambient light, no harsh direct source"
• Specify detail preservation: "skin texture visible in bright areas, fabric detail in highlights"
• Bright clothing on bright background = highest risk: force contrast between them
• For windows/sky in frame: "correctly exposed window view, no blown sky, interior and exterior balanced"
• SDXL negative prompt: "overexposed, blown highlights, washed out, white clipping"