Instruction Lens Score: Your Instruction Contributes a Powerful Object Hallucination Detector for Multimodal Large Language Models

Multimodal large language models (MLLMs) have achieved remarkable progress, yet the object hallucination remains a critical challenge for reliable deployment. In this paper, we present an in-depth analysis of instruction token embeddings and reveal that they implicitly encode visual information while effectively filtering erroneous information introduced by misleading visual embeddings. Building on this insight, we propose the Instruction Lens Score (InsLen), which combines a Calibrated Local Score with a Context Consistency Score that measures context consistency of the object tokens. The proposed approach serves as a plug-and-play object hallucination detector without relying on auxiliary models or additional training. Extensive experiments across multiple benchmarks and diverse MLLM architectures demonstrate that InsLen consistently outperforms existing hallucination detection methods, highlighting its effectiveness and robustness. The code will be publicly available.

Multimodal large language models (MLLMs) often describe objects that do not actually appear in an image, which reduces their reliability in real-world applications. Existing hallucination detection methods mainly rely on visual features or external models. We discovered that instruction embeddings inside MLLMs implicitly contain reliable visual information and naturally suppress hallucinated objects. Based on this finding, we propose Instruction Lens Score (InsLen), a training-free hallucination detector that combines calibrated visual evidence with context consistency from instruction embeddings. InsLen can be directly applied to different MLLMs without additional training or auxiliary models. Experiments on multiple benchmarks and models show that InsLen consistently outperforms existing hallucination detection methods and improves the reliability of multimodal AI systems. Our findings suggest that instruction embeddings provide a valuable and underexplored signal for improving the reliability of MLLM generation.