Stable Diffusion XL 1.0
image
open-weight
Stable Diffusion XL (SDXL 1.0, July 2023) is a 3.5B-parameter image generation model. Using a two-stage pipeline (base +...
Version: 1.0
Released: 2y 3m 6d ago on 07/26/2023
Architecture
- parameters: 3500000000
- context_length: 1024
- training_data: Trained on LAION image-text pairs with a latent diffusion U-Net (base + refiner stages)
- inference: text-conditioned diffusion
Capabilities
- Generates photorealistic 1024 images with minimal text prompt.
- Excels at vibrant colors, fine detail, and complex scenes.
- In preference tests, users favored SDXL outputs over other open models.
Benchmarks
- User Preference: Preferred open model
Safety
- Includes content filters to block NSFW and harmful prompts.
- Open model may require additional safety layers in deployment.
Deployment
- regions: private
- hosting: self-hosted (8GB GPU)
- integrations: Available via Stability API.
Tags
image-generationdiffusionopen-source