Recommendations:
1. CLIP termination layer: 2-3
The core of Lora lies in its CLIP termination layer, which is set to level 3, which can show more details, and my training is also based on level 3. Of course, level 2 can also be used.
2. Weight range: customized accuracy (0.6 to 1, recommended: 0.8)
3. Recommend matching models
For unparalleled results, I highly recommend combining Lora with the lbcrealistic model, which can be found at https://civarchive.com/models/181285?modelVersionId=203467
Thank you for choosing My Lora and welcome your valuable feedback and suggestions. If you have any questions or need further assistance, please feel free to contact me.
I have created a new TG channel, everyone is welcome to join: https://t.me/+kNqrrSTSLoc1ZDhl
Description
FAQ
Comments (1)
Would you like to ask, in this "veil" concept, how to deal with the label in the training set? See the tag in your example diagram does not build its own trigger word or does the AI learn "1girl"? (Or is the training method differentiated?)