LoRA Model created from 50 public images of Lea Seydoux.
Warning: Might produce nudity if not appropiated prompt.
Supports are welcome : If you like my work, I will be glad to receive your donation (https://ko-fi.com/barabasj214).
Special requests: Do you want something in particular or a personal LoRA model? Contact me through Ko-fi (https://ko-fi.com/barabasj214) or Patreon (patreon.com/barabasj214).
Description
Generation process reviewed. Size now has its expected value (~144MB).
FAQ
Comments (8)
576 mb for an sd 1.5 lora? woww man could this be pruned?
edit: i mean it looks great but half a giga for a single lora? :O
@Malessar Wow! You are right...something went wrong with the creation process. I'll checl it. My LoRAs normally are 144 MB....
@Malessar New LoRA published (v2). Now is 144MB as usual. Thanks for your comment.
the v2 still lists l34s3y as the trigger word, did it really change to l34s3y-v2? in a111 it shows the tag as l34s3y simply xD
@Malessar I changed the trigger in the Kohya_ss process... mmmm....
Hi, thanks for your work! I notice that your lora works exceptionally well on both realistic and cartoon ckpt. May I ask how you prepared the training set? Did you blur the background or just use the original image, and do you need to caption the images?
Hi @Islandir , I use Kohya_SS for the generation process. The images I choose are not touched in any way, but, I have some "rules" to define which ones will be part of the training set, and yes, I need to caption every image (I use another Kohya service for this matter). Regards.
@barabasj214 Seems very similar to my workflow, but I can't blend the realistic face features into the cartoon style neatly. Guess I'll have to refine my training set. TY again for reply