Many draft iterations later...

• SPRINKLED PIXEL

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)

• Best result x time = efficiency ratio:

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: 4 (hyper-SD) + LoRA needed)

• Sampler: dpmpp_sde_gpu

• Scheduler: karras

• Steps: 4 (Lightning) + LoRA needed)

• Sampler: euler

• Scheduler: sgm_uniform

• Steps: 30

• Sampler: dpmpp_3​m_sde_gpu

• Scheduler: karras

• Steps: 100

• Sampler: dpmpp_2s_ancestral

• Scheduler: lcm

• Steps: 100

• Sampler: dpmpp_2s_ancestral

• Scheduler: exponential

• Steps: 200

• Sampler: dpmp​p_sde_gpu

• Scheduler: karras

• Steps: 200

• Sampler: dpmpp_3m_sde_gpu

• Scheduler: karras

• Steps: 200

• Sampler: euler_ancestral

• Scheduler: exponential

• Steps: 200

• Sampler: dpmpp_2s_ancestral

• Scheduler: exponential