Tattoo identification remains a challenging problem due to the high variability and complexity of tattoo designs. Recent studies have shown that combining visual features with captions generated by multimodal large language models (MLLM) can improve retrieval accuracy by incorporating semantic information beyond purely visual representations. In this work, we extend prior research by evaluating several MLLMs for generating tattoo descriptions. We focus on integrating MLLM-generated textual embeddings with visual features, ensuring a consistent evaluation framework. We analyze how differences in descriptive richness and verbosity among MLLMs influence retrieval results, providing insights into the role of text generation style in multimodal tattoo identification. Experimental results show that MLLM choice and verbosity significantly affect identification performance, with concise yet semantically informative descriptions yielding the best results. To the best of our knowledge, this is the first work to provide a comparative study of multiple MLLMs for tattoo identification, highlighting the importance of balancing semantic detail and verbosity in multimodal retrieval.