We provide scripts for: Generating images across 8 base models (including the detailed prompts). Computing S2-CLIP features for the generated images. Calculating model distances based on the extracted features. Usage Description: Generate Images Use the code under generate-images-script/ and run the command from generate-images.sh to generate images across all models. Compute S2-CLIP Features With the generated images, run the command from compute-s2-clip-features-for-folders.sh to extract S2-CLIP features for every image and model. Output: .npy feature files. Compute Model Distances Once all S2-CLIP features are computed, use compute-model-distances.sh to calculate the distances between models. We use HuggingFace’s fine-tuning script for SD2.1 and Kohya’s SD-Script for fine-tuning FLUX. https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image.py https://github.com/kohya-ss/sd-scripts The dataset for fine-tuning include: PCB: https://huggingface.co/datasets/bghira/photo-concept-bucket LAION-Aesthetics: https://laion.ai/blog/laion-aesthetics/