What problem does it solve? AI image models cannot reliably review their own output, so generated images often ship with extra fingers, distorted faces, garbled text, wrong Vietnamese diacritics, leftover watermarks, or style drift. This Skill runs a separate, evidence-based quality review before an image is used in a SEOSONA Flow pipeline. ## Core Features & Use Cases - Visual checklist review: Scores hands, faces, in-image text, watermarks, style match, composition, and thumbnail legibility as PASS/FAIL with one-line reasons. - Text-integrity check: OCRs all visible text and compares it character-by-character against the expected string, flagging dropped characters, garbled glyphs, wrong diacritics, bad line breaks, and edge clipping. - Actionable verdicts: Returns ACCEPT, REGENERATE (with one concrete prompt fix), or OVERLAY (reserve space and render real vector text instead of trusting the model to spell correctly). - Use Case: After generating 30 product images in one style, run each through this quality gate to catch the image where the headline lost a Vietnamese tone mark, and route it to a text-overlay fix instead of a costly regeneration. ## Quick Start Review this generated image against my brief and expected text, then give me a verdict of ACCEPT, REGENERATE, or OVERLAY with the single most important fix.
