This is my second attempt at creating a LoRa using the combined style of Crypine and Zhu that I always use on my images. I’ve reduced the dataset and selected a more manageable epoch. However, it might still be very unstable, so use it at your own risk.If you’re wondering why this is a new LoRa rather than an updated version of the previous one, the answer is: I don’t frigging know, I have no idea how to do it.

Metaldragorzz merged

This is my second attempt at creating a LoRa using the combined style of Crypine and Zhu that I always use on my images. I’ve reduced the dataset and selected a more manageable epoch. However, it might still be very unstable, so use it at your own risk.If you’re wondering why this is a new LoRa rather than an updated version of the previous one, the answer is: I don’t frigging know, I had no idea how to do it.

metaldragorzzMergedV1_1.safetensors

V1.1

This LoRA is a personal experiment designed to merge the aesthetic of two of my favorite styles (Tian Zhu &amp; Crypine) into a single one. Trained on top of Illustrious. v1.1Second (and more thoughtful) attempt. I've reduced the dataset and manually tagged 67 images (never again). I also picked a more balanced epoch to keep shit stable.Works best around 0.7 to 1.0Still a bit unstable because I'm a noob v1.0My very first raw test using 195 AI-generated images with default trainer settings. A bit overcooked, high artifact risk, and heavily rigid. Use at your own risk.Best at weight 0.9 to 1.0

Reduced dataset, fixed epochs and steps, hoping it would run better. Not sure tho.

Files with SHA256: a02b364bc32763e3e6f766833a7a03d00842ebb9d5deceaa49d461f16a11fbee