YFG Piranesi [Flux] - V1.0 API Documentation

Original images used for training were uncropped and include the full text on the scanned artifact. The LoRA is somewhat overfit and will usually win in prompting. However, the subjects will usually be faithful to the general idea even if the model will insist on the frame and text as the original posters. This is Epoch 16 and a total of about 7500 steps at a Network DIM of 64 to capture the extremely fine details and lines in the originals. I will likely make another weaker version. All images were captioned only with the trigger word.