I was desperate to do something good and just did what I could,this loRA is aimed at stylizing your requests for 80s-90s anime I doubt that lora succeeded, it is far from ideal, not all third-party lora work with it perfectly I think this is due to the fact that I had a fairly small dataset, as before only you decide what and how to do with it, after all SD is a field for experimentation I will not recommend an exact value,use this loRA in the range of 0.3 to 1 If anyone has a desire to improve this loRA, please contact me in comments and I will give you a dataset, as well as the initial settings, in general you can easily do something similar, in my case I used 130 art in the style of old anime that were generated with midjourney

80'sFusion

In general, thanks to people in the comments, I decided to redo the style as the first version was really quite overtrained, in the second version I added 50 more arts and reduced the number of steps per epoch, as well as the number of epochs, in general, as I think loRA generates easy arts like people and cars slightly better, but it now works better with other loRAs, now became easier to generate characters, as well as the cars stopped turning into a Nissan at low weights, in general you can see in the art section

Files with SHA256: 43eb2c015f6f116af8030bb9b14e0be7ed7f695daa0830ff1412344ed89fbfad