Replies: 2 comments 1 reply
-
Yep. Get 50-200 images in a style, tag everything in them, and don't create an activation tag. Increase the resolution too, around 768. You can also set the text encoder lr to 0. Generally, styles are harder to get right compared to characters or concepts, but you can ask around for more advice. |
Beta Was this translation helpful? Give feedback.
0 replies
-
That is, I need to add an image with different content, but in the same style, and write in the tags what is happening in the image, but write nothing about the style? |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I’ve been using Lora_Trainer.ipynb for a long time, but I never understood how to train lora for drawing styles. Is it possible to do this using Lora_Trainer.ipynb?
Beta Was this translation helpful? Give feedback.
All reactions