ｄｏｌｌｘ ｓｔｙｌｅ

ｄｏｌｌｘ ｓｔｙｌｅ

This Lora is trained on the style of AI artist @darkestdollx and creates a hyperwave type of image heavy in pinks and colorful spectrums with glitch and vaporwave vibes. It does not exactly reproduce their style and is intended as a fun tribute to their work. This LoRA was trained for 40 epochs using the prodigy optimizer on 24 images at 1280 x 1280 with a batch setting of 4 and a gradient accumulation of 6. This is different from my normal training settings of 4 x 2. Images were captioned with GPTV using my art captioning prompt and captions were manually edited for succinctness. Loss charts were different than I expected with the different training settings than I usually use. After reviewing test images i chose to share the final epoch 40. Images in the accompanying post were creating with comfy UI and the Proteus Run Diffusion model with the tempest model as a refiner stage. Experiment with Clip Skip -2 for more coherent images if using the Proteus run diffusion model.

Have fun and follow @darkestdollx on twitter!