- This topic has 15 replies, 5 voices, and was last updated 5 months ago by
.
-
Posts
-
My first anime (non-realistic) Lora. From the anime TV series of 2000, aka NieA under 7.
https://en.wikipedia.org/wiki/NieA_7
Thanks gaver. It came out much better than I expected considering I used screen-caps. AI or the base model I trained it on improved the quality. I only used one full body shot but AI was able to perfectly "fill in the missing pieces", legs, feet, profile shots, etc.
Considering this is done on sd1.5 I think it came out extremely well! I like how the images turned out, plus elves 🙂
Thanks Legolas! It took about an hour to train with 1.5 and I really don't have the patience to do it with bigger images in SDXL.
BTW Niea is an alien not an elf, a low ranking under-seven class. That's why the other aliens pick on her.I found a workaround to the problem. It's called img2img with flux. I am using no prompt whatsoever.
It seems to have fixed hands and plants too 🙂
You should put together a guide with your tips and methods and post it for points.
Many would like a structured path to follow - especially given your results.
Maybe co-author with @Frank22
🙂Yes Flux is hugely better with hands but not quite 100% yet, with too many and not enough fingers still occurring. Still though we shouldn't complain. There are now some people training Flux Loras but I think the sys requirements needed for training puts it out of the range of nearly everyone.
Perhaps a massive LLM is the only way to get the results we see with Flux, but that's pretty useless if the model isn't useful for normal people with normal PCs. Some of the claims around Flux accessibility seems a lot like gaslighting. I recall a time when most games and aps were easily accessible to almost everyone.
Thanks Adam. I can't do much because I don't have any of those accounts to login but I did make something... err...disturbing.
- You must be logged in to reply to this topic.
