Generate WCAG-style accessibility descriptions for images at standard or detailed level, via Vision.describe() with Claude or OpenAI.