Problems and Proposed Solutions When a model has learned to focus too much on the specific characteristics of the training data, rather than generalizing to new situations. 2.Overfitting Solution 1 Dreambooth used prior-preservation loss, and the ratio of prior-preservation is never easy to determine. Solution 2 It is training data that caused the overfitting. Thus we use a subset of the training data to train an overfitted model, select the previous checkpoint which and use it to generate images by prompt for a single word. These images can be placed in the regular training data according to the word frequency ratio, and the subset of the which that caused the overfitting can be removed, and then retrain the model from the very beginning.