Good morning! I have a question regarding training with Dreambooth. I used the same dataset of 180 p
Good morning! I have a question regarding training with Dreambooth. I used the same dataset of 180 pictures of myself and trained using sd_1.5 and Deliberate as the base, with the same parameters. However, for each case, I generated the class dataset (photo of man) with its corresponding model. The results are as follows: when using the model trained with SD1.5, my face doesn't come out very well. When using the model trained with Deliberate, my face still doesn't come out very well. However, when I do the following checkpoint merge, the result is really good: (Primary: Deliberate, Secondary: Custom_SD1.5, Tertiary: SD_1.5, Add Difference, M=0.5).
In other words, I add the difference between my model trained with SD_1.5 and the original 1.5 model to the model I want to use as a base (Deliberate), and this is much better than fine-tuning the Deliberate model with my dataset. Does anyone have any idea why this is happening?
In other words, I add the difference between my model trained with SD_1.5 and the original 1.5 model to the model I want to use as a base (Deliberate), and this is much better than fine-tuning the Deliberate model with my dataset. Does anyone have any idea why this is happening?
