This is the first time I train concept Lora, comments are appriciate
Recommend Prompt: You can try red lips and skindentation if the face not look mature enough
The difference between 3 LoRA versions are training different methods and datasets. V1 is trained on Usual Dreambooth, PixAI is trained on PixAI's New Lora Maker, SeaArt trained on their Lora Training. Mix is like its name, the mix of all three.
LoHA is trained with V1's dataset, but with LoHA training method and it gave kinda different result. LoCon has an improvement of dataset from V1. And A1111 is now support for LyCORIS so you no longer need to install extensions to use LoHA version
Since this Lora has too many versions, I think I would recommend you using 1 of 3 best versions: LoCon, Mix3 and LoCon2. The comparasion between them is at gallery.
The newest version is for SDXL, you can ignore it if you are using SD15
There are some images I left with no prompt, some are from PixAI and some are from tensor.art. You can download and you use "SD Prompt Reader" to read prompt
Description
FAQ
Comments (13)
Works with Pony diffusion?
I just take a try and yep, it can
I just have realeased new version trained with Pony Model, you can try it.
@Rendai Wow! Thanks! Deff gonna try!
Why CLIP skip 2?
I don't know how to explain but CLIP Skip 2 makes my images better
@Rendai It shouldn't be neccessary for SDXL from what I understand. There are a lot of loras and checkpoints that require clip skip 2 because they're trained on NovelAi's model which requires clip skip 2. But if you aren't training on something based on NovelAi's SD 1.5 models, then clip skip 2 just makes it a bit more annoying to use the lora without any real gains.
Sao phiên bản SDXL nhìn ngon hơn bản Sd1.5 thế nhỉ
Cơ bản thì SDXL nó đã ngon hơn SD15 rồi bác, nên Lora train và sản phẩm nhìn sẽ lứng hơn SD15
cool!!
you planned to release a Pony version man ?
Yes, I do! And I'm still testing it
Sorry for long delay, I just have released Pony Version!