Stable Diffusion XL 1.0

Model design and access

A two-stage workflow can pass noisy base latents to the refiner. The card also describes an image-to-image refinement approach. These are pipeline choices within the SDXL release. The base card lists limitations in readable text, faces and precise spatial compositions.Stable Diffusion XL 1.0 official model documentation

The current API documentation retains SDXL 1.0 under the stable-diffusion-xl-1024-v1-0 engine. It lists supported width and height pairs, including square and wide formats.SDXL 1.0 text-to-image API

The SDXL-based Stable Image Core service

Stability describes Stable Image Core as an image service that uses an enhanced version of SDXL. The cited announcement does not identify a separate public training checkpoint. Core is covered here as a hosted SDXL-based service. Its service name does not identify the unchanged SDXL 1.0 weights.Stability AI: Stable Image Core uses enhanced SDXL

The current Core endpoint accepts a text prompt, aspect ratio, negative prompt, seed and style preset. It returns a 1.5-megapixel image as image bytes or encoded JSON. These service controls are separate from the settings exposed by a local SDXL pipeline.Stable Image Core official model documentation

Practical use

Practical guidance. SDXL remains useful where a local workflow already has compatible adapters and control tools. Add the refiner only when it improves the intended output; another stage also adds memory movement and inference work. Inspect small details and lettering after all refinement and upscaling steps.