Visual Puns from Idioms: An Iterative LLM-T2IM-MLLM Framework
This work addresses the challenge of creating and understanding visual puns for applications in AI creativity and multimodal AI, though it is incremental as it builds on existing models and methods.
The paper tackles the problem of generating visual puns from idioms by developing an iterative framework that coordinates LLMs, T2IMs, and MLLMs for automatic generation and evaluation, resulting in a dataset of 1,000 visual pun images and showing that MLLM choice is the primary performance driver with GPT achieving the highest accuracies.
We study idiom-based visual puns--images that align an idiom's literal and figurative meanings--and present an iterative framework that coordinates a large language model (LLM), a text-to-image model (T2IM), and a multimodal LLM (MLLM) for automatic generation and evaluation. Given an idiom, the system iteratively (i) generates detailed visual prompts, (ii) synthesizes an image, (iii) infers the idiom from the image, and (iv) refines the prompt until recognition succeeds or a step limit is reached. Using 1,000 idioms as inputs, we synthesize a corresponding dataset of visual pun images with paired prompts, enabling benchmarking of both generation and understanding. Experiments across 10 LLMs, 10 MLLMs, and one T2IM (Qwen-Image) show that MLLM choice is the primary performance driver: GPT achieves the highest accuracies, Gemini follows, and the best open-source MLLM (Gemma) is competitive with some closed models. On the LLM side, Claude attains the strongest average performance for prompt generation.