In this version i have reworked 80% of the dataset, including more accurate descriptions. more than 50% of the images i cut myself from 4 movies in 1080p quality. according to my tests the model is now more similar to the original. but i didn't have time to train the model on my machine and i used civitai training.

Training Data:

500px - 1000px = 5 pics

1000px - 1500px = 139 pics

1500px - 2000px = 40 pics

2000px - 2500px = 13 pics

Total Pics = 197

197 Pics x 50 Epoch = 9,850 Steps Total