Stable Diffusion

Stable Diffusion is the foundational open source image generation model that has enabled an entire ecosystem of tools, platforms, and fine-tuned models. Released by Stability AI in 2022 with open model weights, it democratised access to text-to-image generation by allowing anyone with sufficient hardware to generate images locally without subscription fees or usage limits.

The significance of the open weights release cannot be overstated for the AI image generation ecosystem. It enabled the Hugging Face and CivitAI communities to produce thousands of fine-tuned models, LoRAs, and specialised variants optimised for specific styles, subject types, and artistic genres. The combination of a capable base model and a vibrant community of fine-tuners has produced a depth of creative options that no proprietary closed model can match.

For local deployment, Stable Diffusion requires a sufficiently capable GPU — typically a modern NVIDIA GPU with at least 4GB VRAM for smaller models, and 8-12GB for larger variants. Interfaces like AUTOMATIC1111, ComfyUI, and InvokeAI provide feature-rich front-ends for the models without requiring programming knowledge. Users who set up local Stable Diffusion have truly unlimited generation capacity with no per-image cost, complete privacy, and full control over prompts and outputs.

The trade-off is technical complexity. Setting up Stable Diffusion locally requires following technical instructions, managing Python environments, and troubleshooting GPU and software issues. For non-technical users, this barrier is significant. Consumer-friendly tools like Midjourney, Adobe Firefly, and Canva AI are more appropriate.

Stability AI has released newer models including Stable Diffusion 3 and SDXL, which produce significantly improved image quality compared to the original SD 1.5 release. The FLUX model family, released by Black Forest Labs (founded by former Stability AI researchers), has become the preferred base model in many community workflows.