I know, it has been done over and over... but not as a LoRA yet. So here it is, hope Aitrepreneur will be happy with the results of this model.
Use kohya-ss/sd-webui-additional-networks (github.com) in auto1111 to load the LoRA model. You can combine it with other LoRA for blended poses.
Description
Multi-resolution model from the same dataset as v1.0 but leveragind 384x384, 448x448 and 512x512 resolution of the same 16 source images.
I also disables 8bitsadam because I discovered that models tend to look better when not using it... Multi-Resolution models are supposed, in theory, to produce better faces and details when the subject is smaller in the overall picture.
So you are the judge if v2 is better than v1.
FAQ
Comments (5)
The best training king
I wish you would provide a simple guide on your training method
Hey, how do you decide to use the subject's name as a trigger or not? What are the caveats if the subject name is already triggered by the original model?
Hi thank you for your work, its really helpful for beginners like me.
I want to ask you, I did everything as its said in "copy generation data", but I don't get christina's face.
when I add <lora:christinaHendricks_v20Mr:1> in the prompt, I get it but its very deformed.
(Checkpoint: v1-5-pruned.ckpt [e1441589a6])
Am I doing something wrong or I just need to keep generating photos and eventually get a better result?
Thank you very much
https://www.instagram.com/wowval_/ pleaseeee
This has to be the best model here. Any chance to do Diane Guerrero or Kylie next?