Edit: v2 released trained on 100 images with a unique trigger word.
Embedding of Natalie Portman.
I get best results with other human-centered checkpoints. Using the base SD 1.5 sometimes gives wrinkles and demon looking eyes so be wary.
Description
Trained on 100 images for 15,000 steps
New unique trigger word "natyport"
I think it's a more realistic looking model
FAQ
Comments (3)
This is a great embedding! What was the learning rate set to for this training? What did you use as the initializer token? How much variation was there across all of the images? Full body, face only, side views, etc?
Combo of everything, truthfully I'm not sure why this worked so well. It's the only one i've made that's not only been any good at all, but somehow really good. I think I grabbed the learning rate from some reddit post i cant find
@lapalla100 Well nice work! This is a fabulous training, thanks for sharing!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.