"You already made a character LoRA why did you make a TI?"
Honestly i'm curious to see which people would prefer more.
In my opinion the LoRA produces content more original to the anime it self while the TI preserves the look of the character better but not the style.
Trigger word: lucy_heartfilia_ti
Link to LoRA: https://civarchive.com/models/6266/lucy-heartfilia-lora
As usual copy the file to stable-diffusion-webui/embeddings
Trained on the same data as the LoRA minus the custom tags.
Description
FAQ
Comments (4)
very well done and inspiring stuff, thanks for letting us know your training data .
in TI, do you actually made a txt file next to each image ? i heard we can do without it.
thanks for your TI anyways :)
no i don't caption the pictures in TI causes it just works somehow.
Hello, i love your work, Could you tell me what settings you use to train the TI?
- How many images do you use?
-How many vectors?
-Do you use "style_filewords" or "Subject_filewords"?
Another question is what model do you use to train anime characters?
Thank you 👉👈
those are a lot of questions you be better of just joining the Project AI discord instead.
I used a google collab that someone else made so i can only answer three out of those four questions.
roughtly 200 images, 20 vectors and nai-wd model.