A versatile lightning model.
(VAE is not baked in)
Recommended VAE: sdxl_vae.safetensors
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix/tree/main
Basic positive prompt
highly detailed, intricate texture, cinematic lighting, vibrant colors
Basic negative prompt
low quality, monochrome, underexposed, low saturation, watermark
LoRA > Offset
https://civarchive.com/models/137511/sdxl-offset-example-lora
clip skip 2
Recommended sampler: DPM++ SDE Karras / Euler A
Sampling steps: 4-6
CFG: 1.5-2.5
Resolution: 896x1152
This model is for educational and non-commercial purposes, and not to cause harm, injury or violate laws.
Description
Version 2 brings general improvements in prompt coherence and leans more towards semirealism.
Some prompt tricks below for using this model for realistic and anime images:
Realism: analog film, photorealistic,
Anime: anime, digital art
(No need for negative prompts unless you specifically need them.)
Recommended settings:
Sampler: DPM++ SDE Karras // Euler A
(Euler A has smoother touch, I'd suggest using it for realism)
Sampling Steps: 4 - 6
CFG 1.5 - 2.5
Resolution: 896 x 1152 // 768 x 1024
AfterDetailer for facial aesthetics. (not required for face portrait)
Upscaling: optional