Published: February 28th 2023, 5:32:08 am
Here are some of my best generations with this LORA! You can see the progression as I tweak and re-train the LORA. I threw in a really funny blooper at the end 😂
Example prompt:
masterpiece, best quality, 1girl, 1boy, arms hanging, sword  swallowing position, kneeling, throat bulge  <lora:sword_swallowing_position3-000004:1>
Negative prompt: arms behind back, easynegative, dark-skinned male
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 44167900, Size:  640x512, Model hash: f303d10812, Model: AOM3A1, Denoising strength:  0.5, Clip skip: 2, ENSD: 31337, ControlNet Enabled: True, ControlNet  Module: none, ControlNet Model: control_openpose-fp16 [9ca67cc5],  ControlNet Weight: 1.5, ControlNet Guidance Start: 0, ControlNet  Guidance End: 0.25, Hires upscale: 2, Hires steps: 10, Hires upscaler:  lollypop
I've found these controlnet OpenPose pictures (see last images) help to guide it a lot. See last image for my controlnet settings. ZIP file has images with prompts, though they're almost all identical to this one
I've also attached an earlier LORA, sword_swallowing_position-000004 -- this had more training data but might be messier