And before anyone says my previous pictures are too plastic, here's a close-up with the same model.
And before anyone says my previous pictures are too plastic, here's a close-up with the same model.
14 Replies
Of course, it could be counterbalanced with negative prompts, fiddling, or using Adetailer, but I have enough evidence that two female Lora in one prompt work fine if they get separate tokens. The eye color is not different, but one of the weaknesses of the SD is the color...
In fact, in the latent space, the Lora features are somewhat blurred, but you can still see the difference between the two models. Here is the unique output of my other model.
you know they did bleed insanely i think. they look like same person
thats quite a jawline
Hello! I would like to ask, I want to train a scene lora, mainly for photography background replacement, what kind of lora should I train, scene lora or style lora or concept lora?
Yes, because they are both part of the woman class, and if I don't use regional prompts, there is a big risk. Unfortunately, the Hires Fix mixes the two women even more, absolutely not prepared for two unique individuals from the same class. However, if they both had the ohwx woman keyword, then there would be an average face from a mix of the two, rather than one dominating.
true
Regional prompter can work wonders, and I didn't perfect the image because it was more about composition for me.
This is a higher CFG (9), it seems worth increasing for more Lora?
Hi, folks! I would like to ask, how to continue training your model from last checkpoint in kaggle dreamboth notebook?
stop the gui
delete other files
give it as a base model
or best is
upload that model into hugging face
as diffusers and give its path
I don't understand why you deleted my posts, and if it was the model's name that was the problem, it would have been enough to delete that one line (that's why I wrote it separately). Anyway, one less experience for the rest of us.
sent a private message to you. yes it was about the name and images
someone already used Microsoft kosmos for captioning?