This was made for the art style contest. This Lora was trained on various art styles and artists whose work was both bright and had a medium-heavy painting style that also included details. The data set also included generated art that I used to emphasize the style I was going for. The images were trained at 768 and aspect ratios of 768 so the best results will be had using those. The best way that I found to use this Lora is to not use a negative prompt and to use the Radiant V4 model using 50 steps, but feel free to experiment with it. specific training details. trained for 20 epochs selecting the 10th epoch using batch size of 2 and bucketing littiting bucketing to 8 or less buckets. 240 images trained for 20 times per epoch. using cosine training with a 10 warmup rate and wadam training and Lr of 0.0001 and Text encoder Lr of 0.00005, network rank of 128 and network alpha of 80. The model was trained using <a rel="ugc" href="https://civarchive.com/models/4509/radiant?modelVersionId=38663">radiant.v4</a> as the base and seems to work best with it.

Files with SHA256: 37cb09efbaf8d295db2496071937084c861255ba329316642881c6f5a5f08dfb