Score-Control for Hallucination Reduction in Diffusion Models

Hugging Face Daily Papers Papers

Summary

This paper introduces Variance-Guided Score Modulation (VSM) to reduce hallucinations in diffusion models by controlling score function smoothness, achieving up to ~25% reduction while maintaining image quality.

Diffusion models have emerged as the backbone of modern generative AI, powering advances in vision, language, audio and other modalities. Despite their success, they suffer from hallucinations, implausible samples that lie outside the support of true data distribution, which degrade reliability and trust. In this work, we first empirically confirm previously proposed hypothesis that score smoothness causes hallucinations in Image Generation diffusion models and provide a density-based perspective. We further formalize this notion by linking the hallucinations probability mass to lipschitz constant of the learned score function. Motivated by this, we introduce a Variance-Guided Score Modulation (VSM) strategy that controls the score Jacobian, in turn reducing score smoothness and better approximating the ground truth score that decreases hallucinations. Empirical results on synthetic and real-world datasets demonstrate that our approach reduces hallucinations (up to ~25%) while maintaining high fidelity and diversity, providing a principled step toward more reliable diffusion-based image generation. We also propose two benchmark datasets with extreme semantic variation for systematic hallucination evaluation. Code and Datasets are publicly available at https://github.com/bhosalems/VSM.
Original Article
View Cached Full Text

Cached at: 06/04/26, 03:41 AM

Paper page - Score-Control for Hallucination Reduction in Diffusion Models

Source: https://huggingface.co/papers/2606.00377

Abstract

Variance-Guided Score Modulation reduces hallucinations in diffusion models by controlling score function smoothness through Jacobian modulation while maintaining image quality.

Diffusion modelshave emerged as the backbone of modern generative AI, powering advances in vision, language, audio and other modalities. Despite their success, they suffer fromhallucinations, implausible samples that lie outside the support of true data distribution, which degrade reliability and trust. In this work, we first empirically confirm previously proposed hypothesis that score smoothness causeshallucinationsinImage Generationdiffusion modelsand provide a density-based perspective. We further formalize this notion by linking thehallucinationsprobability mass tolipschitz constantof the learnedscore function. Motivated by this, we introduce aVariance-Guided Score Modulation(VSM) strategy that controls thescore Jacobian, in turn reducing score smoothness and better approximating the ground truth score that decreaseshallucinations. Empirical results on synthetic and real-world datasets demonstrate that our approach reduceshallucinations(up to ~25%) while maintaining high fidelity and diversity, providing a principled step toward more reliable diffusion-basedimage generation. We also propose two benchmark datasets with extremesemantic variationfor systematichallucination evaluation. Code and Datasets are publicly available at https://github.com/bhosalems/VSM.

View arXiv pageView PDFGitHub0Add to collection

Get this paper in your agent:

hf papers read 2606\.00377

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.00377 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.00377 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.00377 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

PARALLAX: Separating Genuine Hallucination Detection from Benchmark Construction Artifacts

arXiv cs.CL

This paper reveals that much of the reported progress in LLM hallucination detection is due to benchmark construction artifacts, where ground-truth answers are embedded in prompts, allowing a simple text-similarity baseline to achieve near-perfect scores. Through a large-scale controlled evaluation, the authors show that most methods perform near chance under proper controls, except for supervised probes on upper-layer hidden states such as SAPLMA and their proposed DRIFT.