Trained on the visual image series 'Tellurion' by Matt Rhodes :
Female concepts taken from mainly Dave Cheung , blvefo9 and unidentifiedsfm , safe for work stuff only because thats all we can make on Tensor anyways.
Also added 'first person view' hands for lulz . See training data for further info.
Learning rate 0.004 using 'cosine with repeats scheduler'
Rank 32 and alpha 16
Trained using 80 images x 5 repeats x 100 epochs = 40,000 training step using Chroma HD
Using batch size 4 , with 40,000 / 4 = 10,000 computation steps
Training data (mild suggestive themes) : https://huggingface.co/datasets/codeShare/lora-training-data/blob/main/tellurion_chroma.zip