Read the damn description!!!
This is a NSFW LoRA for Life is Strange. More SFW stuff to come in the future. Please note that this is my FIRST LoRA. Feel free to give me tips if you encounter problems or have any suspicions about which configured parameters are causing the error.
I recommend “CyberRealistic Pony v8” or “2DN-Pony” as a checkpoint.
V1
Known issues and limitations:
A small chance of logos/watermarks appearing in the generated images despite negative prompt “watermark, logo, text, words, fonts”.
The prompt “public hair” was trained to “pubic hair” due to a spelling mistake. 🙃
Improvements in the future:
Background optimization, such as “max's dorm room” or “chloe's bedroom”.
Improvement of “sitting on ___” or “lying on ___” results.
More training on SFW data for better clothing prompting. Blender 3D models serve as a source for this. 😉
Addition of Max from Life is Strange: Double Exposure.
Description
V1
Known issues and limitations:
A small chance of logos/watermarks appearing in the generated images despite negative prompt “watermark, logo, text, words, fonts”.
The prompt “public hair” was trained to “pubic hair” due to a spelling mistake. 🙃
Improvements in the future:
Background optimization, such as “max's dorm room” or “chloe's bedroom”.
Improvement of “sitting on ___” or “lying on ___” results.
More training on SFW data for better clothing prompting. Blender 3D models serve as a source for this. 😉
Addition of Max from Life is Strange: Double Exposure.
FAQ
Comments (5)
Even though it is your first Lora
It is very good.
You should be very proud of it
Thank you for your nice comment! 🙂
It took me 11 versions and a lot of testing to achieve this result.
You can use rank 32 for dim and alpha its more than enough.
Thanks for your advice! 🙂
As I understand it, the network dimension increases the details and dynamics of the model. I set the network dimension so high because this is not just a character LoRA, but several characters and poses. In other words, this is a LoRA with a high dynamic range and a comparatively large data set.
I'm actually planning to increase the amount of training data and also increase the network dimension as a result. Are you sure that a network dimension of 32 is enough for this LoRA?
Maybe I can test it out. 🤔
This is the 11th version I have trained. The previous configurations/versions were not very good.
This version with a network dimension of 32 and a network alpha of 1 is the best so far.
@TensorRipper You can check my loras almost all are rank 32, the artist styles I aways use over 100 imgs datasets, and they work pretty good.