Stable Diffusion 3.5 🧠 Model Open source

What it is

Stable Diffusion is Stability AI's family of open-weight text-to-image models. The current version is Stable Diffusion 3.5 (October 2024) in three variants: Large (the most capable, for 1 megapixel images), Large Turbo (four-step generation) and Medium (designed for consumer hardware). SDXL and SDXL Turbo remain available as legacy models. Stability's 2026 news has centred on Stable Audio 3.0 and funding, with no new image model announced.

Key features

• Open weights you can download, fine-tune and run offline
• MM-DiT architecture with three text encoders for better prompt adherence than SDXL
• ControlNets (blur, canny, depth) for SD 3.5 Large to guide pose and composition
• Styles from photography to line art and 3D
• The largest community library of LoRAs and fine-tunes, mainly built on SDXL and SD 1.5

Where you can use it

Locally in ComfyUI, InvokeAI (SD 1.5, SDXL, SD 3.5 Medium and Large) and Fooocus (SDXL-based). Leonardo.ai still offers SDXL 1.0 and several SDXL-based models. Stability also runs it on its own Developer Platform API, in the Stable Assistant web app and through cloud partners.

Pricing and rights

The weights are free under the Stability AI Community License for individuals and organisations with under $1M in annual revenue, including commercial use, and you own your outputs. Larger companies need an Enterprise licence. Stable Assistant uses credits (6.5 credits per image); we could not confirm a current monthly plan price, so price_from is empty. The SD 3.5 reference code on GitHub is MIT-licensed.

Who it is for

Hobbyists, technical artists and studios who want full local control, custom styles and no per-image costs.

Verdict

Stable Diffusion is still the backbone of the open, local image ecosystem thanks to its tooling and fine-tunes. As a model it is dated in 2026: prompt following and text rendering lag newer families, and no successor has been announced.