Verified: 4 months ago
SafeTensor
Type | |
Stats | 218 80 |
Reviews | (21) |
Published | Jul 19, 2024 |
Base Model | |
Training | Steps: 825 Epochs: 10 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | childrens_book_illustration |
Hash | AutoV2 C1DFCB1D9E |
This model is for childrens' book illustrations.
Trigger word: childrens_book_illustration
For better results you can try the keyword sharp.
Improvements from v2.0 to v2.1
Improved output quality by upscaling low resolution images and passing them through a cartoon restoration ESRGAN model
Improved prompt adherence by adding new images of more subjects
Removed a low quality image
Fixed issue where portrait images would generate as lanscape ones, rotated. This was caused by incorrect orientation of some of the source images (Rotation was marked in its EXIF data, but that is not respected during training)
Improvements from v1.0 to v2.0
Recreated, expanded and recaptioned entire dataset