SearchGen-Bench

Evaluating search-grounded image generation across knowledge boundaries

751evaluation prompts
22knowledge domains
12failure modes
10score components
Submit Here Paper Project Page 🤗 HF Paper Code 🤗 Data
Loading benchmark data...

Click any column header to sort. Domains and failure modes are multi-label: group counts overlap. Coverage is shown because missing results are zero-filled or excluded according to the canonical model-specific policy.

Cite SearchGen-Bench

If you found SearchGen-Bench useful, please cite us at: