Bring your favorite virtual assistant waifu to life!

This LORA model was trained with 98 images (SFW and NSFW) of Samantha. It was trained with the trigger phrase "SamanthaSamsung", although you shouldn't need to use it if the LORA is loaded into the prompt with the weight assigned.

This LORA model can produce both SFW and NSFW images.

I recommend using a LORA weight between 0.7 and 0.9. Anything lower than 0.7 causes the results begin to diverge from Sam's likeness. 1.0 and higher can produce over-saturated results with bad faces and deformed eyes.

Your experience may vary, but I typically create renders with CFG between 3 and 6 for a more 2D likeness of Sam. 7 and higher makes her appear more 3D. I don't recommend going above 10.

Recommended ckpt/safetensors base models:

I recommend using either Dreamshaper, Neverending Dream, or Art & Eros, as they seem to work very well with this LORA model. You can use the base SD 1.5 model that the Samsung Sam LORA was trained with, but it is harder to work with to get the results you want.

Known Issues:

You may have issues with Sam's eyes in poses that aren't looking directly at the viewer, or in poses where she is turned around and looking back. You can try to brute force this with a large number of iterations, but most times I end up having to manually redraw and correct the eyes.

In the training images, I had to edit out the Samsung logo on her black jacket since it produced nonsense letters in the beta versions of these LORA models. You'll have to edit in the logo yourself if you want it in.

How to use LoRA's in auto1111:

• Update webui (use git pull like here or redownload it)

• Copy the file to stable-diffusion-webui/models/lora

• Select your LoRA like in this video

• Make sure to change the weight (by default it's :1 which is usually too high)