Stable Diffusion
Also known as: SD, SDXL, Stable Diffusion 3, SD3, Stable Diffusion 3.5
Stable Diffusion, first released in 2022, became the foundation of an enormous community ecosystem. Because the model weights (the trained parameters) are publicly available, anyone can download and run the model locally, which led to an explosion of fine-tuned variants, tools like ComfyUI (a node-based visual workflow builder) and AUTOMATIC1111, and community hubs like Civitai where creators share custom model variations.
The current generation as of 2025 is Stable Diffusion 3.5, available in Large (8B parameters), Large Turbo (faster), and Medium (designed to run on consumer hardware). SD 3.5 introduced a Multimodal Diffusion Transformer architecture that replaced the older U-Net design, significantly improving text rendering inside images and handling of complex prompts. Stability AI also offers API access through their Developer Platform if you want to skip local setup.
The practical reason builders care about Stable Diffusion specifically is flexibility and cost: there are no per-image charges when running locally, you can fine-tune the model on your own data using techniques like LoRA (a method for adding a style or subject without retraining the whole model), and you can deploy it privately in your own infrastructure. The trade-off is setup complexity and the need for adequate GPU hardware.