This research introduces FoodMLLM-JP, a pioneering multimodal large language model specialized for Japanese recipe generation from food images.

• Developed by fine-tuning existing MLLMs with a custom Japanese recipe dataset containing 306K image-recipe pairs
• Achieves superior performance in generating culturally appropriate Japanese cooking instructions compared to general-purpose MLLMs
• Successfully produces recipes with accurate ingredients and procedures aligned with Japanese culinary traditions
• Demonstrates the potential of specialized MLLMs to address domain-specific challenges in food understanding

This advancement significantly impacts gastronomy by enabling automatic dish recognition and recipe creation that respects cultural cooking nuances, supporting both culinary education and dietary management applications.

FoodMLLM-JP: Leveraging Multimodal Large Language Models for Japanese Recipe Generation