Hi everyone,this LoRA / LyCORIS was my first try in preparing a dataset and run a training locally on my M1 Max. The training was done with the original SDXL base model.use "humanpup" and/or "pup mask" within the prompt and a CFG of 0.6 to 0.8 (depending on model you are using).The preview image was created by an combination of text2image and inpainting with invokeai.

HumanPup

For this version i switched from OneTrainer to Kohya and doubled network rank and alpha. Training was done with the Prodigy Optimizer and (in comparision to v0.1) 30% larger dataset.Some words about my workflow: I rarely do soley text2image generation. Usually i create a rough draft with text2image and switch to inpainting, upscaling and inpainting of details again.

Files with SHA256: 9b635c28da1aea7611f700516cb0c03e7c7650c792c45aef08a1a278a11c40a8