Understanding How MLLMs Describe Artworks Using Token Activation Maps
For researchers studying interpretability of MLLMs, this work provides a method to probe visual grounding in art description, revealing token-specific patterns and hallucination tendencies.
The authors use Token Activation Maps (TAM) to analyze how Multimodal Large Language Models (MLLMs) ground their descriptions of artworks in visual regions. They find that grounding varies by token type, and that MLLMs are more accurate at artist attribution than title prediction, with hallucinations more common in the latter.
Multimodal Large Language Models (MLLMs) describe artworks with remarkable fluency, yet the visual reasoning behind their outputs remains opaque. When an MLLM names a style, identifies a subject, or recognizes an iconographic symbol, does it ground each claim in the relevant region of the canvas, draw on an undifferentiated visual signal, or rely primarily on textual priors? We study this using the Token Activation Map (TAM), which produces, for each generated token, a heatmap isolating the visual evidence specific to that token from prior-context interference. Applying TAM to a curated set of paintings spanning multiple periods and genres, we analyze grounding patterns across five semantically distinct token categories: common visual objects, style descriptors, metadata, iconographic tokens, and affective expressions. We find that visual grounding varies substantially with token semantics. We further show that MLLMs attempt to identify artworks and artists, achieving higher accuracy in artist attribution than in title prediction, where hallucinations are more frequent. Finally, we compare TAM with SAM~3 open-vocabulary segmentation. To ensure reproducibility, we release our code, experimental configurations, prompts, and qualitative results on the project page at https://nicolafan.github.io/tamart/.