wangkanai / sdxl-fp8

Stable Diffusion XL (SDXL) Model Repository

This directory is reserved for Stable Diffusion XL models and related components.

Repository Status

Current Status : Empty / Placeholder

This directory structure is prepared for future SDXL model storage but currently contains no model files.

Intended Contents

When populated, this directory would typically contain:

SDXL Base Models

• sdxl-base-1.0.safetensors - Base SDXL model (6.94GB)
• sdxl-refiner-1.0.safetensors - Refiner model (6.08GB)

Text Encoders

• CLIP-G and CLIP-L text encoders
• OpenCLIP-ViT-bigG encoder

VAE

• SDXL VAE for latent encoding/decoding

Optional Components

• LoRA adapters
• ControlNet models
• IP-Adapter models

Model Information

Stable Diffusion XL is a state-of-the-art text-to-image generation model featuring:

• Larger UNet backbone (2.6B parameters)
• Dual text encoders for better prompt understanding
• 1024x1024 native resolution
• Two-stage pipeline (base + refiner)

Hardware Requirements (When Populated)

SDXL Base :

• VRAM: 8GB minimum, 12GB+ recommended
• Disk Space: ~14GB for base + refiner
• Memory: 16GB+ system RAM

SDXL with Refiner :

• VRAM: 12GB minimum, 16GB+ recommended
• Better quality with two-stage generation

Usage Example (Reference)

from diffusers import DiffusionPipeline
import torch

# Load SDXL base model (when available)
pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16"
)

pipe.to("cuda")

# Generate image
image = pipe(
    prompt="a beautiful mountain landscape",
    num_inference_steps=50
).images[0]

# Load refiner (optional)
refiner = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16"
)

# Refine image
refined_image = refiner(
    prompt="a beautiful mountain landscape",
    image=image
).images[0]

License

Stable Diffusion XL is released under the CreativeML Open RAIL++-M license, which permits commercial use with certain restrictions.

Official Resources

• SDXL Base Model
• SDXL Refiner
• SDXL Documentation

Note : This is currently an empty placeholder directory. Download SDXL models from Hugging Face and place them here for local storage and usage.