2-Pass Workflow for Pony Diffusion base model
(might work for SDXL models too)
Sample images generated with:
Parameters:
upscale latent values are good at range 1.2-1.6 depends on image complexity
combining two clips also improve details (empty prompt is good too)
image quality and generation speed is good at 1024x768 resolution
to preserve checkpoint style sometimes is good to set LoRA clip strenght to 0
some models works well without score tags at all, which may improve hard concepts
Other Features:
Controlnet (mistoLine)
Soft Inpainting (left bool checkbox)
Img2Img (right bool checkbox)
FreeU, PAG