This is a LoRA for Lewis Griffin from American Guy 😉
It was trained on ~130 pictures with pretty rough prompt pruning.
Most of the training images were ( nude 😳) so clothing is pretty tough. I wanted to try and keep the style mostly the same for this version, but next version will most likely have more images.
The images I uploaded were ran thru 1.5x and 1.7x upscale with no other alterations at all. No VAE, no in-paint, no photoshop. I have a feeling with a little push you can do some better ones with not a whole lot more effort.
Trained on the excellent AnyLoRA model by Lykon.
I generate them using AnyLoRA and use the In-Painting model as well for touch ups.
I also recommend using Automatic1111's highres fix as well, even at low values because this LoRA doesn't like to draw pupils?? <- ✋😤🤚
I plan on updating it a little bit if I get around to it, I accidentally used the baked vae version on AnyLoRA instead of the training model.
I use it at 0.8 strength but I'm sure full is probably fine too. I'm not a master proompter yet so I do like to have it a little weaker for *creativity*
Yes, obviously I know the name of the show and the character. Lewis and American Guy are some cheeky little jokes.
Description
First Version
FAQ
Comments (7)
*Lois
Yeah, it's Lois Griffin and it's also Family guy, not American Guy.
Could actually be a strategy to dodge googling and therefor bad eyeballs. Maybe a good idea.
@32Bitshifter That's a fair point.
here's the trick to get her black dot eyes:
black eyes::2, orange hair::2, short hair::2
and in the negative prompts add these:
blue eyes, orange eyes, brown eyes, dark skin, long hair
after many images, i just set in the negative what had been generated the most and added in the positive what definitely has get generated. and it works absolutely wonderful.
unfortunately, i used a too high resolution and it generated a "loli-lois" and "loli-meg"(yes, meg too xD) multiple times.
EDIT: added an example image
EDIT2: loras for clothes seem to trigger her black "dot eyes", because everytime i used these loras, those dot eyes got generated. but for now, it's more trial and error
Try "solo" instead of "1girl" when using a lora (and maybe embedding). It nearly always gives a single subject in my experience vs 1girl.
Doesn't seem like an issue for your samples but also remember to merge prompts that effect the person such as "loisprompt, 30 year old," to "30 year old as loisprompt," if you produce multiple subjects. Gotta remember a computer and a human read things differently so gotta play to know how that checkpoint works best as I've not used that one. Hope that helps you a bit.
@WellTheWasSomething i really thank you for your tips. they will be helpful in the future. i'm pretty sure of it.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.