CRANIAL ORIGAMI STYLE
This LoRA was trained (with consent) on 36 images by the legendary master of prompts @CranialOrigami. It does not exactly duplicate his numerous styles, it is more of a creative tribute to his vision that you can play with. I selected 36 of his images that had a similar vibe. Using Gigapixel I upscaled them to max height of 2048px with noise reduction all the way up to 100. They were captioned using GPTV via API with a special prompt I worked out specifically for image captioning. Captions were then edited down to ~155 token length. They were trained for 40 epochs with a batch size of 4 and a gradient accumulation of 2 using the prodigy optimizer. After testing with various base modes i chose epoch 39 to share. Juggernaut and Tempest.artistic work very well with this model. I prefer the dpmpp_2m_sde_gpu and dpmpp_3m_sde_gpu with kerras scheduler but the classic euler_a works just fine as well. Make sure to follow @CranialOrigami on twitter!