SDXL

Fast and flexible. Dual-CLIP text encoding (CLIP-L + CLIP-G) with UNet denoising and classifier-free guidance.

• Developer: Stability AI
• License: CreativeML Open RAIL++-M
• HuggingFace: stabilityai/stable-diffusion-xl-base-1.0

Variants

Model	Steps	Size	Notes
sdxl-turbo:fp16	4	5.1 GB	Ultra-fast, 1–4 steps
dreamshaper-xl:fp16	8	5.1 GB	Fantasy, concept art
juggernaut-xl:fp16	30	5.1 GB	Photorealism, cinematic
realvis-xl:fp16	25	5.1 GB	Photorealism, versatile
playground-v2.5:fp16	50	5.1 GB	Artistic, aesthetic
sdxl-base:fp16	25	5.1 GB	Official base model
pony-v6:fp16	25	5.1 GB	Anime, art, stylized
cyberrealistic-pony:fp16	25	5.1 GB	Photorealistic Pony

Defaults

Defaults are per model:

• Resolution: 1024x1024, except sdxl-turbo:fp16 at 512x512
• Guidance: 7.5 (sdxl-base, realvis-xl), 7.0 (juggernaut-xl, pony-v6, cyberrealistic-pony), 3.0 (playground-v2.5), 2.0 (dreamshaper-xl), 0.0 (sdxl-turbo)
• Scheduler: ddim (sdxl-base, juggernaut-xl, realvis-xl), euler-ancestral (dreamshaper-xl, pony-v6, cyberrealistic-pony, sdxl-turbo), edm-dpm-pp-2m (playground-v2.5). --scheduler accepts ddim, euler-ancestral, uni-pc, and edm-dpm-pp-2m (Playground v2.5 only).

Recommended Dimensions

Width	Height	Aspect Ratio
1024	1024	1:1 (native)
1152	896	9:7
896	1152	7:9
1216	832	19:13
832	1216	13:19
1344	768	7:4
768	1344	4:7
1536	640	12:5
640	1536	5:12

Using non-recommended dimensions will trigger a warning. All values must be multiples of 16.

Example

SDXL Turbo: 4 steps, seed 88:

mold run sdxl-turbo:fp16 \
  "A vibrant street food market in Bangkok at night, \
  neon signs, steam from woks, bustling crowd" \
  --seed 88

Negative Prompts

SDXL uses classifier-free guidance; negative prompts have a strong effect:

mold run sdxl-base:fp16 "a landscape" -n "low quality, blurry, watermark"