People wearing wet clothes.
Trained on 256 samples, both men and women.
V 2.5XL-alpha:
Initial version for SDXL with no model specific tuning for learning
V 2.3:
Works best with Deliberate.
On realistic models has tendency to crop heads. Use 'cropped', 'top crop' and 'head out of frame' in negative prompt.
Describe what kind of clothes character is wearing on may generate strange ultra shine ones. Works best on denim.
Description
Initial version for SDXL trained with same data and captures as 1.5 versions
FAQ
Comments (10)
Did you accidentally tag this as SDXL, Deliberate is a 1.5 model iirc
If you check the samples it says Model: sd_xl_base_1.0
@escheresque I suppose he was mentioning the description of the lora, which was talking about the previous versions (that were not for SDXL)
@escheresque Ahh gotcha, yeah I was just going by the 'Works best with Deliberate.' in the description
Sorry forgot to update model generic description
Is it easy to train loras on SDXL? I'm still trying to guess (I don't have the hardware to do so) because the dual text encoder makes me think it could be way harder than the previous SD's
Process when using Kohya_ss is equivalent, tho I think it trains both text encoders similarly. This my alpha version uses about similar parameters as with 1.5 version, no any specific tunings yet. (Adafactor as optimiser instead of AdamW)
Love your lira. Any plan to make a new version for comfyui. It cannot create the wet effect so need a lora asap.
These should work with comfyUI also. I am using that tool these days. Sorry been off for a while but hoping to get back to AI art this winter..Meybe even some new Lora/model if I have time
it works very well