# Unmasking LVLM Hallucinations

_New research introduces the HalluScope benchmark, revealing textual priors as the main driver of LVLM hallucinations. A new framework, HalluVL-DPO, uses preference optimization to improve visual grounding._

**Published:** 2026-04-24
**Source:** https://www.startuphub.ai/ai-news/ai-research/2026/unmasking-lvlm-hallucinations

---

Large Vision-Language Models (LVLMs) exhibit impressive capabilities but remain susceptible to generating outputs not grounded in visual input. A critical question has been the relative contribution of vision backbone limitations versus language dominance to this hallucination problem. New research from Khayatan et al. introduces the [HalluScope benchmark](https://arxiv.org/abs/2604.21911v1), a novel tool designed to dissect the factors inducing these visual grounding failures.

## Excessive Textual Priors Fuel Hallucinations

The analysis conducted using the HalluScope benchmark reveals a significant finding: LVLM hallucinations largely stem from an over-reliance on textual priors and background knowledge. This is particularly evident when information is introduced through textual instructions, suggesting that the language component's learned associations can override or misinterpret [visual](/ai-news/ai-research/2026/rubric-driven-dpo-for-visual-tasks) cues. This insight challenges previous assumptions and provides a clearer target for mitigation strategies.

## HalluVL-DPO: Grounding Language with Preference Optimization

To address the identified issue of instruction-induced hallucinations, the researchers propose HalluVL-DPO. This framework employs preference optimization, fine-tuning existing [LVLMs](/ai-news/ai-research/2026/perceptio-spatial-grounding-for-lvlms) using a carefully curated dataset. The training process guides the model to favor visually grounded responses over those prone to hallucination. Initial results demonstrate that HalluVL-DPO effectively reduces targeted hallucination failure modes while maintaining or even improving performance on other benchmarks and core visual capabilities.

---

Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.