In the dataset here: https://huggingface.co/datasets/xingxm/SVGX-Core-250k you describe that the img_analysis column is "Analytical description of the image's visual features." how was this generated? which model / prompt did you use?