Retrain here on civitai, with expanded set (260), using the same tagging (extended text made with joy caption with more than 150 tokens), LR 0.0002 (as suggested by @Joschek), 64/64. I am not sure on the epoch (here is e 15, but lower epochs produce good results in the few samples also). Start with strength at 0.8 and lower if you use other LoRAs. It might be overtrained, so if you want, ask me to upload other epochs (I kept some more)
