The captioning LLaVa AI was given specific prompt instructions to caption descriptions using a concept object-in-context strategy which focused on the jockstrap itself first, such as material, positioning, angles and other materially important attributes. Then other elements in the picture would be a secondary description priority. So, when using this LoRA a specificity such as waistband size, attributes about leg straps, pouch, etc., should work better than without.