We release an early report and analysis on failure modes of GPT-4V and LLaVA-1.5, later released as HallusionBench.