Trained 90s pornstar Jenna Jameson as on my potato PC. It worked out pretty good, and i wanted to share this with you. Giving back for all the knowledge and work i got from this community 👊.
I'd love to see your creations using this Lora! If you like this model, please click the Like button and leave a review, it really helps for visibility! 🤗
v.1.5 changes
😍 Works fine on weight of 0.7 to 1
😍 Cleaned up training images. Removed some bad quality ones, removed dust, scratches, blemishes and skin imperfections.
😍 Better captions
v1.0
😍 Works just fine on weight of 1
😍 Plays very well with other lora's and styles
🤞 On full body shots you need to have some luck, because there are some issues with rendering the mouth, close ups work like magic tough!
😟 Need to clean up my pics because blemishes pop up where we don't want them
😟 Overall needs some love on the training pics, something for the next version
😟 Train a little more/further next time, i have the feeling this could be much better
Example prompt
full face portrait of jenna jameson a beautiful 25 year old blond woman model with big long hair wearing a dress, <lora:jenna_jameson_v1.0:1>Add any negatives you want.
Thanks for watching and reading! 🤗❤️
Description
FAQ
Comments (4)
I praise you for using the name of the subject as the trigger word. It is such a simple thing, and almost nobody does it, for the life of me I can't understand why.
Because it's not necessarily a good idea. If SD already knows that subject by name (e.g. Taylor Swift) it will inject a random amount of the information about that person that it already has, usually with worse results.
In this case, although it's unlikely that SD1.5 itself will have much training on Jenna Jameson, some trained Checkpoints may do. Much better to use 'JJameson' (or similar) as the trigger word, no risk of confusion.
@MegaWalrus Yeah, good point, never really given it much thought. Can i change that after i trained or do i need to train again?
@Dennuz666 AFAIK it would need retraining. In this instance though the LoRA works well with all the Checkpoints I've tested it with (albeit mostly merged ones rather than trained ones), so I think it's OK. Just a tip for the future I guess.