Type | |
Stats | 439 1,014 3.5k |
Reviews | (44) |
Published | Sep 7, 2024 |
Base Model | |
Training | Steps: 2,105 Epochs: 10 |
Usage Tips | Clip Skip: 2 |
Trigger Words | young asian male, |
Hash | AutoV2 1C5DBB7AAF |
East-asian male with androgynous face.
Trained with over 200+ images from my previous pony-based lora.
I am using Flux.1-Dev BNB NF4
NOTICE: Try generating 8~10 images first with 10~12 steps. First few images will be very blurry.
after that, it'll be a-okay.
Due to the training data, and my personal preference some images with female outfits might give your result ....breasts. if so, consider lowering the base weight + using certain prompts such as "man, male focus, masculine, handsome, etc..." and use adetailer with a lora as a prompt to get the desired result. This issue is very easily avoidable.
Preferred weight is around 0.4~0.7, lower weight, more controlled output.
VER_2.0!
I had no idea flux hated large sample size lol
ver.2 used 30+ training data with handwritten t5 encoder captions. trained on 768^2 res.
more focused on the face/hairstyle + aesthetic of the photo.
Recommended setting
Euler/DDIM, 20~30 steps,