Simple Portraitphoto-Style based on 190 Images
Description
20 Images, 20 captioned, same 20 uncaptioned
Strength of 0.7 seems a good choice
FAQ
Comments (4)
Hey ! So you find a way to make good loras for flux with larger datasets (as you were recommending 20/30 pictures max) ? How many steps ? :) Thanks for your work
I just used the rapid mode on flux lora training onSite with joycaption, still in testing mode :D And i have to release the Loras since i only can use the bnb4 Model locally which gives totally different results.
Actually, really great! It gives so good results with only 2000 steps. Any plans on larger training like 5000 or 10000 steps?
Hihoo ;-) Interesting - you wrote: Trained with 20 Images, 20 captioned, same 20 uncaptioned - why did you use another 20 uncaptioned?
Greetings from a fanboy of your models ;-)