V0.2 updated
Single character dataset with masked watermark.
Trained to 2250 steps
This is my very first attempt at making a LoRA for Flux Dev.
Trained using https://github.com/kijai/ComfyUI-FluxTrainer
AdamW - Cosine with Restarts
?? Dim
?? Network
Learning rate 3e-4
1600 steps (200 epoch) on 8 random images of jollyjack's art.
Seems to work best at 0.7 strength.
Description
Changed the dataset to only be of single characters. Masked watermark.
Trained to 2250 Steps (about 250 Epochs)
I did a 3000 step version, but it was overtrained and kept putting heels on everyone.
Details
Downloads
99
Platform
SeaArt
Platform Status
Available
Created
10/5/2024
Updated
10/9/2024
Deleted
-
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.