Yes, sref but also tuning, pluse blending and other img2img approaches are definitely great way to break prompts. In my experiments I don't use figurative prompts anymore, but e.g. inputing poems or abstract ideas - in this case new aesthetics emerge. Btw I am often working with my present --all (will write later), which includes all the MJ versions to the same prompt. Sometimes earlier versions deliver even better results.