Stable Diffusion 4.
The open-weights image model — the foundation of the entire generative-art ecosystem.
What it is.
London-based open-source pioneer. Stable Diffusion 4 shipped in 2026 with improved text rendering and prompt adherence. The model that built the entire downstream ecosystem — ComfyUI, Automatic1111, every LoRA marketplace, every fine-tune on Civitai. Stability survived a 2024 leadership crisis and shipped under new CEO the company's leadership.
Where it fits.
Wherever the operator needs control, customisation, or sovereignty. Studios building proprietary style models. Enterprises with data-residency requirements blocking hosted APIs. Game studios generating thousands of asset variations against a fixed art direction. The base layer most creative pipelines still depend on.
- Largest open ecosystem — LoRAs, ControlNet, ComfyUI
- Self-hostable for full data sovereignty
- Strongest fine-tuning toolchain in image generation
- Out-of-box quality trails FLUX and Midjourney
- Workflow complexity rewards specialists, punishes newcomers
Frequently asked.
Is Stable Diffusion still free?
Yes for non-commercial use. Commercial deployment under the Stability Community Licence is free below $1M annual revenue, paid above.
Why use Stable Diffusion over FLUX?
Ecosystem and customisation. SD has thousands of LoRAs, ControlNet adapters, and ComfyUI workflows. FLUX leads on raw quality.
What hardware do I need to run it locally?
SD 3.5 Medium runs on 12GB VRAM. SD 4 Large needs 24GB. Cloud rental from Replicate or RunPod starts at $0.50 per hour.
The photorealism frontier — built by the Stable Diffusion founders, now powering xAI Grok.
The image-model reference point — still the aesthetic benchmark in 2026.
The commercially-safe creative model — trained on licensed Adobe Stock, wired into every Creative Cloud app.