i used images generated by the first lora to train the second.
it was very hard to get good images, so i took the best ones, photoshopped all the imperfections and inconsistencies out, and used those to train this new model. This lora should be way easier to use and give much better results.
