Trained with 28 images, using just the standalone keyword is sometimes enough, but using more specific descriptions yields better results. The writing works sometimes, using something like "writing says "name"" should trigger it.
hololivecheerup, midriff, skirt, holding pom poms,