Trained with 7 images, using just the standalone keyword is sometimes enough, but using more specific descriptions yields better results. Due to the dataset being very limited in variety, I recommend using lower weights than my usual models. Around 0.7-0.8 should suffice.
amiaryllisbloo,