The experiment with Gemini 2.5 Pro aimed to generate P5.JS code based on three images, focusing on replicating specific patterns. While it successfully recognized basic shapes, it struggled with compositional arrangement, leading to inaccurate final outputs. Future attempts may benefit from clearer prompts and simpler models for better results.