A family of image-generation models from Stability AI, including SDXL and Stable Diffusion 3.5 variants. Explore model selection, hosted or self-hosted inferenc
Last reviewed: 2026-10-02
Stable Diffusion names a model family rather than one application or subscription. Stability AI provides image models for deployment on your own infrastructure and access through hosted services. Record the model version and the service you use; two products displaying the Stable Diffusion name need not use the same model or settings.
Within the 3.5 family, Stable Diffusion 3.5 Large is the full-sized variant, Stable Diffusion 3.5 Large Turbo is a distilled variant designed for fewer sampling steps, and Stable Diffusion 3.5 Medium is a smaller model. SDXL is a separate model generation. Choose the documentation, weights, and workflow for the exact variant rather than treating these names as interchangeable.
The official SD3.5 repository supplies reference inference code and documents separate downloads for the model and text encoders. It also lists ControlNets for SD3.5 Large. Check compatibility before carrying an adapter or control workflow from SDXL or another variant into a new setup; availability for one model is not a promise for every model.
For an apparel concept experiment, start with a brief and references your team is permitted to use. Compare a small set of prompts against the same review criteria, keep the model and settings with each output, and have a designer inspect pattern continuity and garment details. Label generated images as concepts until the team has approved their intended use.
Pricing: There is no single Stable Diffusion subscription price. Self-hosting has hardware, compute, and operating costs, plus the requirements of the exact model license. Hosted services set their own usage charges. Check Stability AI's current licensing page and your chosen provider's pricing before estimating a project; model access does not make hosting or every commercial use free.
This is a family-level guide covering the distinction between SDXL and the Stable Diffusion 3.5 Large, Large Turbo, and Medium variants. Consult the model card for the exact checkpoint you plan to use; the list is not a claim that every host exposes all variants.
Do not assume compatibility. Start from the chosen variant's documentation and confirm that each model file, adapter, and control component supports it. Record the working configuration before comparing results.
Begin by checking whether an existing model can address a small, defined brief. Record where it fails. Consider customization only after you can describe the missing behavior, the permitted training material, and how you will review the results.