For best outputs:
Sampler: DPM++ SDE Karras, DPM++ SDE Karras.
Steps 4, cfg 1 ~ 1.5, CLIP skip 2.
VAE: sdxl_vae.safetensors (highly recommended).
Pos prompt: starting your prompt with "Masterpiece, extremely detailed, best quality" actually improves quality. This model also makes use of the LoRa "sdxl_offset", which improves contrast and perceived image quality. Weight to your taste.
Neg prompt: this one will do "(low quality, worst quality:1.2), lowres, multiple views, border, (grayscale, grainy), (watermark, signature, text:1.1), underexposed, overexposed, oversaturated, low saturation".
This is a semi-realistic model. It cannot create perfect realism neither anime, but instead it can create beautiful outputs quickly (feel free to check my example runs to see what it can do).
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
For best outputs:
Sampler: DPM++ SDE Karras, DPM++ SDE Karras.
Steps 4, cfg 1 ~ 1.5, CLIP skip 2.
VAE: sdxl_vae.safetensors (highly recommended).
Pos prompt: starting your prompt with "Masterpiece, extremely detailed, best quality" actually improves quality. This model also makes use of the LoRa "sdxl_offset", which improves contrast and perceived image quality. Weight to your taste.
Neg prompt: this one will do "(low quality, worst quality:1.2), lowres, multiple views, border, (grayscale, grainy), (watermark, signature, text:1.1), underexposed, overexposed, oversaturated, low saturation".
This is a semi-realistic model. It cannot create perfect realism neither anime, but instead it can create beautiful outputs quickly (feel free to check my example runs to see what it can do).
