Stable Diffusion

As one of the foundational models in the AI image generation landscape, it can be run locally on consumer hardware, accessed through Stability AI's official API, or used via numerous third-party platforms that have built products on top of its architecture.

The latest iteration, Stable Diffusion 3.5, uses a Multimodal Diffusion Transformer (MMDiT) architecture available in three variants: Large (8.1 billion parameters) for maximum quality, Large Turbo (distilled for speed) for faster generation without significant quality loss, and Medium (2.5 billion parameters) optimized for performance on consumer-grade hardware.

This tiered model structure allows users to balance quality, speed, and local hardware requirements.

Advanced editing capabilities go beyond text-to-image generation. Inpainting allows specific areas of an existing image to be edited and regenerated while preserving the rest.

Outpainting extends the canvas beyond the original image boundaries, generating new content that seamlessly matches the existing style and composition. Depth-to-image transformations use depth maps to guide generation, preserving spatial structure while changing visual content.

The open-source nature of Stable Diffusion has made it the foundation for a vast ecosystem of fine-tuned models, community modifications, and specialized applications.

Users can run the base model locally for complete privacy and unlimited generation, use fine-tuned variants for specific aesthetics or domains, and integrate the model into custom workflows through the extensive community tooling that has developed around it.