Trained with 30 images using DPM++_2M for SD3.5L Base model for 6000 steps over 20 epochs
Recommended LoRa strength: 0.5
CLIP skip: 1
Link to loss chart + training data:
You can download the training data as a zip file here:
Captions were created using JoyCaption Alpha One :
JoyCaption Notebook:
Keyword origin
The keyword was selected by running a training image through this notebook:
One of the similiar results according to CLIP model in the text_encoding notebook was "art by Brian Sum", so I googled that and behold "Brian Sum" was actually a guy who draws robots! You can find his creations here:
I did add 4 images of Brian Sum's works into the robot LoRa, bringing up the total from 26 images to 30.