A collection of various character loras (some of you may be familiar with my posts on CivitAI) trained at 1024 resolution on Pony Diffusion V6 XL and intended to be used with said model as well as derivative models such as AutismMix SDXL.

These are trained using DoRA targetting the linear and conv layers of the unet while freezing the text encoder. All datasets were re-labeled and some were expanded with additional images to improve quality.

Training was done using manually-gathered datasets and Derrian Distro's training scripts on 1x 4090. Detailed training data, including trained tags, is provided in the safetensors metadata included with each lora.

These are provided as-is, and YMMV. The user is responsible for any outputs produced using these loras, and outputs may vary depending on the base model used.