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:https://imgur.com/a/blPjv6S

You can download the training data as a zip file here: https://huggingface.co/datasets/codeShare/lora-training-data

Captions were created using JoyCaption Alpha One : https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-one

JoyCaption Notebook: https://colab.research.google.com/github/camenduru/joy-caption-jupyter/blob/main/joy_caption_jupyter.ipynb

Keyword origin

The keyword was selected by running a training image through this notebook: https://huggingface.co/datasets/codeShare/fusion-t2i-generator-data/blob/main/Google%20Colab%20Jupyter%20Notebooks/fusion_t2i_CLIP_interrogator.ipynb

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: https://www.artstation.com/sum .

I did add 4 images of Brian Sum's works into the robot LoRa, bringing up the total from 26 images to 30.

Brian Sum Mech LoRa - e16