CHARM: New Multimodal Graph Foundation Model Enables Zero-Shot Transfer Across Domains
Researchers have introduced CHARM, a multimodal graph foundation model designed to perform zero-shot transfer across different graph domains without requiring downstream adaptation, according to a paper published on arxiv.org on July 29, 2026.
According to the paper, existing graph foundation models face limitations: GNN-based models typically require downstream adaptation, while LLM-based graph methods mainly address unimodal graphs or tasks within a single domain. CHARM addresses two key challenges in zero-shot transfer on multimodal graphs. First, models must generalize knowledge from individual modalities while capturing transferable cross-modal relations. Second, without target-domain fine-tuning, node representations remain entangled with domain-specific structures and modality-specific characteristics, according to the researchers.
The CHARM framework replaces isolated raw nodes with hierarchical graph contexts that capture multimodal semantics and cross-modal relations, according to arxiv.org. These contexts map domain-specific node patterns to shared high-level concepts, reducing reliance on target-domain supervision or adaptation. A modality-aware graph context encoder integrates multimodal information with graph structure and converts the resulting representations into graph tokens for a large language model.
According to the paper, experiments show consistent improvements on zero-shot multimodal graph tasks, though specific performance metrics were not detailed in the available excerpt.