Many draft iterations later...
A merge of many, a master of none.
Overview: "Sprinkled Pixel: A versatile, user-friendly, and reliable daily driver for Stable Diffusion XL. This advanced-merge base checkpoint is designed for seamless 'set-and-forget' functionality. It delivers exceptional results with mostly minimal effort."
Basic Recommendation For Potentially Better Results;
VAE = 1st: stabilityaisdxl-vae.safetensors
VAE = 2nd: ponyDiffusionV6XL_vae.safetensors
VAE = 3rd: Fooocus Default (model)
(please feel free to experiment & test with your own favorite VAE)
Use Hyper-SD (4-step) with LoRA model (sdxl_lightning_4step_lora.safetensors) > weight 0.8 or...
Use Lightning (4-step) with LoRA model (sdxl_hyper_sd_4step_lora.safetensors) > weight 1.0
Scale for sharpness ratio: 2/3 is a good start to your control bar or 65% full
Guidance Scale: Start at 1.0 and move up as desired
CLIP Skip: 2 is for best prompt adherence, 1 & 12 have hidden gems
NSFW Model resists NSFW generation, but is easily persuaded
How I get my images?
Performance: Hyper-SD
Steps: 4
Resolution: (1024, 1024)
Guidance Scale: 1.0
Sharpness: 0.0
ADM Guidance: (1.0, 1.0, 0.0)
Base Model: Sprinkle_Pixel_fp16.safetensors
Refiner Model: None
Refiner Switch: off
CFG Mimicking from TSNR: 1.0
CLIP Skip: 2
Sampler: dpmpp_ sde_gpu
Scheduler: karras
VAE: stabilityaisdxl-vae.safetensors
LoRA 1: sdxl_ hyper_sd_4step_lora.safetensors : 0.8
Version: Fooocus v2.5.5
Basic Examples of Other Acceptable Generation Settings;
Steps: 200
Sampler: dpmpp_sde_gpu
Scheduler: karras
Steps: 200
Sampler: euler_ancestral
Scheduler: exponential
