One Model, Many Graphs: Learning over Attributed Graphs across Heterogeneous Modalities with Vision-Language Models

2026-07-21

AI Summary

Existing graph learning approaches struggle to uniformly handle graph data with heterogeneous modalities—such as text, images, or multimodal attributes—often requiring distinct models for each modality, which limits generalization and scalability. This work proposes OMG-VLM, a novel framework that leverages pretrained vision-language models (VLMs) as a universal backbone for graph learning, augmented with a lightweight structure-aware graph adapter. By integrating multimodal node features and graph topology within a single architecture, OMG-VLM eliminates the need for modality- or structure-specific model redesign, substantially enhancing generalization to unseen graphs and novel modality combinations. Experimental results demonstrate that OMG-VLM consistently outperforms state-of-the-art graph neural networks and large language model baselines on node classification and link prediction tasks, exhibiting exceptional cross-graph generalization and modality adaptability.

Abstract

Vision-language models (VLMs) provide a unified representation space for textual and visual information, yet their potential as general-purpose backbones for graph-structured data remains largely unexplored. In practice, attributed graphs exhibit substantial modality heterogeneity: some graphs contain only textual node attributes, others only visual attributes, while still others provide both. Existing graph learning approaches are typically designed for fixed modality schemas, requiring separate models for different settings and limiting scalability and cross-graph generalization. To bridge this gap, we present OMG-VLM (One Model, Many Graphs with Vision-Language Models), a unified framework for learning over attributed graphs across heterogeneous modality schemas. OMG-VLM leverages a pretrained VLM as a shared backbone and introduces structure-aware graph adapters that integrate neighborhood information while remaining compatible with the VLM's native embedding space. This design enables effective learning over text-attributed, image-attributed, and multi-attributed graphs within a single model. Extensive experiments across diverse domains show that OMG-VLM consistently outperforms state-of-the-art GNN- and LLM-based baselines on attributed graph learning tasks such as node classification and link prediction, while exhibiting strong generalization to unseen graphs and varying modality schemas. The source code is available at https://github.com/Jo-eyang/OMG-VLM.

Problem

• attributed graphs
• modality heterogeneity
• graph learning
• vision-language models
• cross-graph generalization

Innovation

• vision-language models
• attributed graphs
• heterogeneous modalities
• graph adapters
• cross-graph generalization