🔍 Structured Defect Grounding (SDG)

Detect localized defects in text-to-image generations and get instance-level feedback — where the defect is, what type it is (🔴 artifact / 🔵 misalignment), why it's wrong, and how important it is (1–100).

Powered by the SDG Detector (P1n3/sdg-detector-grpo), a Qwen3-VL-4B model trained with SFT + GRPO from the paper Where, What, Why, and Importance: Structured Defect Grounding for Text-to-Image Feedback.

Upload a generated image and (optionally) the caption it was generated from, then press Analyze.