Despite strong performance of Multimodal Large Language Models (MLLMs) on multimodal tasks, predicting whether and why an image is persuasive remains challenging, yet no established methodology exists for training MLLMs to reason about visual persuasion or evaluating whether their rationales faithfully support their decisions. To address this gap, we show that diverse teacher-generated rationales, when used for supervised fine-tuning, improve visual persuasiveness prediction. We further introduce a three-dimensional faithfulness evaluation framework covering rationale-to-decision consistency, rationale-to-image groundedness, and rationale-to-decision sensitivity. Applying this framework shows that prediction performance alone does not guarantee faithful rationales, while rationale-to-decision sensitivity is most aligned with human rationale preferences. These findings motivate faithfulness-aware training objectives and scalable rationale supervision for visual persuasiveness evaluation. Our code and dataset will be made publicly available.

Vision-language models (VLMs) have achieved strong performance across diverse multimodal tasks, yet they remain vulnerable to unreliable reasoning. Existing self-correction methods mitigate these issues but typically rely on post-training or carefully engineered feedback, incurring high computational cost. In this work, we revisit this challenge through the lens of emotional cues, asking whether they can activate latent self-correction behaviors in VLMs without additional training. We find that emotional signals serve as an effective trigger for self-correction, encouraging more cautious and reflective reasoning. Motivated by this finding, we propose ESC (Emotional Self-Correction), a training-free self-correction framework. ESC introduces an external verifier that detects potentially incorrect initial responses and injects emotional feedback to encourage model to reflect, and produce a better revised response without additional training. Extensive experiments across safety, hallucination, vision-centric perception, and multimodal reasoning benchmarks show that ESC consistently improves reliability while preserving overall model utility. These results suggest that emotion can function not only as an ability to be recognized, but also as a practical control signal for scalable self-correction in VLMs. We therefore believe that ESC provides a strong foundation for a new reliable human-like, emotion-integrated research direction.

The web is littered with images, once created for human consumption and now increasingly interpreted by agents using vision-language models (VLMs). These agents make visual decisions at scale, deciding what to click, recommend, or buy. Yet, we know little about the structure of their visual preferences. We introduce a framework for studying this by placing VLMs in controlled image-based choice tasks and systematically perturbing their inputs. Our key idea is to treat the agent's decision function as a latent visual utility that can be inferred through revealed preference: choices between systematically edited images. Starting from common images, such as product photos, we propose methods for visual prompt optimization, adapting text optimization methods to iteratively propose and apply visually plausible modifications using an image generation model (such as in composition, lighting, or background). We then evaluate which edits increase selection probability. Through large-scale experiments on frontier VLMs, we demonstrate that optimized edits significantly shift choice probabilities in head-to-head comparisons. We develop an automatic interpretability pipeline to explain these preferences, identifying consistent visual themes that drive selection. We argue that this approach offers a practical and efficient way to surface visual vulnerabilities, safety concerns that might otherwise be discovered implicitly in the wild, supporting more proactive auditing and governance of image-based AI agents.

Vision-language models (VLMs) are increasingly deployed in real-world and embodied settings where safety decisions depend on visual context. However, it remains unclear which visual evidence drives these judgments. We study whether multimodal safety behavior in VLMs can be steered by simple semantic cues. We introduce a semantic steering framework that applies controlled textual, visual, and cognitive interventions without changing the underlying scene content. To evaluate these effects, we propose SAVeS, a benchmark for situational safety under semantic cues, together with an evaluation protocol that separates behavioral refusal, grounded safety reasoning, and false refusals. Experiments across multiple VLMs and an additional state-of-the-art benchmark show that safety decisions are highly sensitive to semantic cues, indicating reliance on learned visual–linguistic associations rather than grounded visual understanding. We further demonstrate that automated steering pipelines can exploit these mechanisms, highlighting a potential vulnerability in multimodal safety systems.

Evaluation resources for persuasive images, memes, ads, and arguments report a number called persuasiveness, but they do not measure one construct. A desk audit of 17 evaluation resources from the vision, NLP, and LLM-safety communities (2016-2026) codes, for each resource, what its effectiveness number operationalizes, whether uncertainty is reported, and whether a model judge could be scoring its own family's output. Five row-traceable findings follow. At least five constructs circulate under the one word: annotated technique presence, annotator judgment, self-reported attitude shift, behavioral or platform-engagement outcome, and LLM-simulated outcome. Measurement discipline divides along community lines: preregistration is standard in the audited social-science experiments, with interval estimates in three of the four, and both are absent from the benchmark lineage, which contains every audited visual corpus. Judge circularity is concentrated in the audited 2024-2026 LLM-evaluation resources, and none of the four exposed designs reports a self-preference or judge-family control. In the one audited resource that measures self-report and behavior jointly, the two signals disagree. And no audited benchmark-lineage resource scores predicted effectiveness as a probabilistic forecast, so an overconfident persuasiveness predictor is invisible to the metrics those resources report. A five-part measurement protocol answers the findings: declare the construct, report uncertainty on headline numbers, score effectiveness predictions with proper scoring rules, control judge circularity, and anchor proxies to behavior where feasible.

Visual persuasion is inherently subjective and personalized, yet existing evaluators typically produce a single generic judgment rather than a viewer-specific one. We reformulate personalized visual persuasion evaluation as a viewer-specific ranking problem, in which an evaluator model's ranking of candidate images for a target viewer is compared against that viewer's own preference ordering. Building on this formulation, we introduce three complementary metrics, Target-profile Ranking Agreement (TRA), Personalized Ranking Gain (PRG), and Profile Identification Score (PIS), that jointly measure ranking quality, the gain from conditioning on a viewer's own profile relative to other viewers, and profile identifiability. We instantiate this framework on the Personalized Visual Persuasion (PVP) dataset, evaluating three multimodal Large Language Model (LLM) families under multiple profile conditions. Across all models, profile conditioning consistently reduces overall ranking agreement, while personalization gains and profile identification remain small and inconsistent. These results indicate that current multimodal LLMs do not yet meaningfully adapt their persuasiveness judgments to individual viewers, motivating more effective profile conditioning and training-based approaches for personalized visual persuasion evaluation.

Graphic design forms the cornerstone of modern visual communication, serving as a vital medium for promoting cultural and commercial events. Recent advances have explored automating this process using Large Multimodal Models (LMMs), yet existing methods often produce geometrically inaccurate layouts and lack the iterative, layer-specific editing required in professional workflows. To address these limitations, we present PosterCopilot, a framework that advances layout reasoning and controllable editing for professional graphic design. Specifically, we introduce a progressive three-stage training strategy that equips LMMs with geometric understanding and aesthetic reasoning for layout design, consisting of Perturbed Supervised Fine-Tuning, Reinforcement Learning for Visual-Reality Alignment, and Reinforcement Learning from Aesthetic Feedback. Furthermore, we develop a complete workflow that couples the trained LMM-based design model with generative models, enabling layer-controllable, iterative editing for precise element refinement while maintaining global visual consistency. Extensive experiments demonstrate that PosterCopilot achieves geometrically accurate and aesthetically superior layouts, offering unprecedented controllability for professional iterative design.

Recent text-to-image (T2I) models generate semantically coherent images from textual prompts, yet evaluating how well they align with individual user preferences remains an open challenge. Conventional evaluation methods, general reward functions or similarity-based metrics, fail to capture the diversity and complexity of personal visual tastes. In this work, we present PIGReward, a personalized reward model that dynamically generates user-conditioned evaluation dimensions and assesses images through CoT reasoning. To address the scarcity of user data, PIGReward adopt a self-bootstrapping strategy that reasons over limited reference data to construct rich user contexts, enabling personalization without user-specific training. Beyond evaluation, PIGReward provides personalized feedback that drives user-specific prompt optimization, improving alignment between generated images and individual intent. We further introduce PIGBench, a per-user preference benchmark capturing diverse visual interpretations of shared prompts. Extensive experiments demonstrate that PIGReward surpasses existing methods in both accuracy and interpretability, establishing a scalable and reasoning-based foundation for personalized T2I evaluation and optimization. Taken together, our findings highlight PIGReward as a robust steptoward individually aligned T2I generation.