Lora created by micvrotom, he deleted it a while ago, with his permission, I reuploaded it
This Lora is basically Croc Artist's style to generate images of the Simpsons characters
The tag serves to reinforce it, but normally it already does so.
For the eyes, use any eye color(for example "Black eyes"), without the eyes tag, they will deform.
Like teeth and hair, you have to define an expression and hair color respectively, since it can be deformed or use a different hair style that doesn't fit.
Use the tags "Marge Simpson, Lisa Simpson, Bart Simspon" in negatives to not generate them, since there are times that they are generated, for example, making someone with long hair tends to generate marge.
Many thanks to micvrotom, for creating it and for allowing me to upload it.
Edit: According to several tests.
With the following settings
Steps:25-35
CFG:7
Sampler :Euler a
Clip Skip: 2
Here in Civitai they generate well with minimal errors, although it will depend on the negative tags of each person.
Description
FAQ
Comments (5)
Hahaha, ironically I can't get it to work xD
Strange.
Do you use Civitai to generate it or an external AI? (Easy Diffusion for example)
If it is in Civitai, in the settings below check the Random box, since custom will keep the image as a template.
And to do what you want, use Booru tags.
Also check if you are using the Pony V6 XL model and the LoRa, since there are times when you hit Remix they do not appear by default.
If it is external, you have to have the Pony V6 XL model in addition to the LoRa, use the booru tags and put negatives that you can copy in the images section.
And in the description there are settings that I have tested and they work well.
If you need help with anything else, tell me.
Thanks bro!, but from the beginning I knew it was my fault, I use a1111, and I wasn't very clear on how to use the XL models, but I already studied a little and finally generated images with your lora without problem, thanks! ^^
Seems like it has issues making homer?
He's not really present in the comic pages that were used to to train this. So he's not in the dataset, or has minimal presence. You might want to train a separate homer lora.