This is an odd one. This embedding works best for cartoon and anime checkpoints, some semi-real checkpoints are hit and miss, but where it works it works well.
The entire concept is to re-create the expressive yet often exaggerated grief of characters from anime and animated properties. The grief of a husband, wife or child crying at the grave of the partner, friend or parents, fearing for the future. Maybe a character turning the corner only to see a monster/creature consuming their co-workers or classmates, let alone maybe fleeing a group of zombies. The exaggeration of concepts find it difficult to translate to the realistic models.
As with realistic models, most cannot grasp the exaggerations or get confused with NSFW scenarios. Which is also what occurs with most checkpoints if the embedding is weighted heavily. Difference being, some checkpoints will tend towards blood and gore, others may ignore the intent totally or go straight into the 'adult'-club variety.
Edit: I had added some trigger words for the embedding but they did not show up so I'll add them here. Grief, Crying, Tears, Fear & Screaming. One or two of these has a big impact with the embedding. However, I'd also suggest based on weighting, adding blood, gore to the negative prompt if possible. Especially for weights over - 1.5