The V2 version has changed the basic model for training, and according to the comments, the anime model of sd1.5 has been used for training. At the same time, better background images were used, totaling 1100 images. The model effect of the V2 version can be presented normally on different checkpoint models, and checkpoint models such as anything, cetus, OrangeMix, ghoatMix, etc. have been tested.
This version has improved image logic, color effects, and stylization, and also has good effects on animated graphics. It is recommended to use it.
Issues to be improved:
Due to the excessive number of training images, the processing of prompt words is relatively sloppy and not handled very meticulously.
2. For wires with the characteristic expression of the Monogatari series, the learning effect of shadow processing is weakened.
Version update V2:
The V2 version has changed the basic model for training, and according to the comments, the anime model of sd1.5 has been used for training. At the same time, better background images were used, totaling 1100 images. The model effect of the V2 version can be presented normally on different checkpoint models, and checkpoint models such as anything, cetus, OrangeMix, ghoatMix, etc. have been tested.
This version has improved image logic, color effects, and stylization, and also has good effects on animated graphics. It is recommended to use it.
Issues to be improved:
Due to the excessive number of training images, the processing of prompt words is relatively sloppy and not handled very meticulously.
2. For wires with the characteristic expression of the Monogatari series, the learning effect of shadow processing is weakened.