🇬🇧 UK · Stable Diffusion (Stability AI)
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: The original open-weight text-to-image model that democratised AI image generation. Released August 2022 by Stability AI (UK). Spawned the open-source AI image ecosystem (LoRAs, ControlNet, fine-tunes, etc.) — and remains foundational.
Front-matter facts
| Field | Value |
|---|---|
| Vendor | Stability AI (London, UK) — originally founded by Emad Mostaque |
| Country / origin | 🇬🇧 UK |
| Recommended for Australian users? | ✅ Yes — open-weight models freely downloadable globally |
| Privacy summary | Weights are open; running them locally = no data leaves your machine; via API (Stability cloud / Replicate / fal / etc.) = vendor’s policy applies |
| Free tier | Open weights are free (CreativeML Open RAIL-M license + some Stability licenses) |
| Paid tiers | Stability AI Membership (paid for hosted access); via API providers (Replicate, fal.ai, Together, etc.) per-image pricing |
| First released | Stable Diffusion v1 — August 22, 2022 |
| Last reviewed | 2026-06-26 |
| Official site | https://stability.ai |
What it is
Stable Diffusion is the original open-weight text-to-image AI model — the model that opened up AI image generation to the open-source community. Released August 22, 2022 by Stability AI, a UK-based AI lab.
Why it matters:
- First widely-usable open-weight image model — anyone could download and run it
- Spawned the open-source AI image ecosystem: LoRAs (specialised model adaptations), ControlNet (pose / structure conditioning), countless fine-tunes, custom UIs (Automatic1111, ComfyUI, InvokeAI, etc.)
- Democratised AI art — at a time when DALL-E 2 was waitlisted and Midjourney was Discord-only
Model generations
- Stable Diffusion v1.4 / v1.5 (Aug-Oct 2022) — the foundation; massive community ecosystem built on these
- Stable Diffusion v2 / v2.1 (Nov 2022) — controversial; community largely stayed on v1.5
- Stable Diffusion XL (SDXL) (July 2023) — major quality leap
- Stable Diffusion 3 / 3.5 (2024) — current closed-source under Stability Membership
- Stable Image (2024+) — Stability’s commercial API offering
- Stable Video (2024+) — video model
- Stable Audio (2023+) — music gen
The community ecosystem
The real power of Stable Diffusion is the community built around v1.5 / SDXL:
- CivitAI — the largest community of fine-tunes, LoRAs, custom models
- HuggingFace — model hosting
- Custom UIs: Automatic1111 WebUI, ComfyUI, InvokeAI, Forge, Fooocus
- LoRAs: tiny model adaptations for specific styles / characters / concepts
- ControlNet: pose / depth / canny edge conditioning
- Inpainting / outpainting / img2img workflows
This ecosystem is unique — no other AI image model has comparable community depth.
What you’d use it for
- Self-hosted image generation without per-call API costs
- Fine-tuned models for specific styles / characters / brand work
- High-volume image generation at low cost (your hardware)
- Privacy-sensitive image work (data never leaves your machine)
- Custom workflows with ControlNet (pose-guided, depth-guided, etc.)
- Specialised use cases (medical imaging, architecture, scientific visualisation) where community has fine-tuned models
- Learning AI art — the open ecosystem teaches you how diffusion models work
When NOT to use:
- For polished consumer experience (Midjourney / Imagen / gpt-image-1 cleaner UX)
- For best-in-class quality with minimal setup (frontier closed models)
- For non-technical users (community ecosystem assumes some technical skill)
How to use from Australia
Run locally (free)
- Need a decent GPU (RTX 3060 12GB minimum for comfortable use; RTX 4090 for fast SDXL)
- Install Automatic1111 WebUI (github.com/AUTOMATIC1111/stable-diffusion-webui), ComfyUI, or InvokeAI
- Download model from CivitAI / Hugging Face (search “Stable Diffusion XL” or fine-tunes)
- Run via web UI; generate locally
- Cost: hardware + electricity only
Use via API providers (cheap, no hardware)
- Replicate — pay per image
- fal.ai — fast, cheap inference
- Together AI — open-weight inference
- Stability AI API — direct from Stability
Use via integrated products
- Canva uses Stable Diffusion / Flux for some Magic Studio features
- Adobe Firefly does NOT use Stable Diffusion (Adobe trained their own)
- Many third-party AI image apps use Stable Diffusion under hood
What it costs
Open-weight, self-hosted
- Free — weights downloadable from Hugging Face / CivitAI
- Hardware + electricity costs
Via API providers (per image)
- Replicate: ~US$0.01-0.05 per image
- fal.ai: similar
- Stability AI API: comparable
Stability AI Membership (commercial use of newer models)
- Creator US$20/mo
- Professional US$50/mo
- Enterprise quoted
How it compares to alternatives
| Capability | Stable Diffusion | Midjourney | Imagen | gpt-image-1 | Flux |
|---|---|---|---|---|---|
| Open weights | Yes (community ecosystem) | No | No | No | Partially |
| Community / ecosystem | By far the largest | Some | Limited | Limited | Growing |
| Fine-tuning / customisation | Best (LoRAs, ControlNet) | Limited | Limited | Limited | Some |
| Run locally | Yes | No | No | No | Yes |
| Polished consumer UX | Limited | Best (web app) | Excellent (in Gemini) | Excellent (in ChatGPT) | Limited |
| Top-end quality | Strong (SDXL+) | Distinctive aesthetic | Best photorealism | Best instruction-following | Strong |
| Best for | Open-source / self-hosted / custom workflows | Artistic | Photoreal + Workspace | Chat-integrated | Customisable photoreal |
For self-hosted, fine-tuned, or community-driven AI image work, Stable Diffusion’s ecosystem is unmatched.
Privacy / data handling
- Self-hosted = data never leaves your machine
- Via API providers — each has their own privacy posture
- Stability AI Membership: standard SaaS privacy
Recent changes
- 2024-26: Stable Diffusion 3 / 3.5 / Stable Image; Stability AI corporate turbulence
- 2024: Emad Mostaque (founder) departed Stability AI
- 2023: SDXL released; major quality leap
- August 2022: Original Stable Diffusion v1 released — foundational moment
Gotchas
- Stability AI as a company has had turbulence — financial issues, founder departure (2024); the OSS ecosystem continues thriving on v1.5 / SDXL independently
- License nuance: v1.5 / SDXL are CreativeML Open RAIL-M (permissive); SD3 / 3.5 are under Stability Membership for commercial use
- NSFW potential — open weights mean users can fine-tune for any content; CivitAI has clear adult-content sections
- For Aussie users, self-hosted Stable Diffusion is a strong choice for privacy / custom workflows
- Community ecosystem is huge — overwhelming for beginners; recommend starting with Automatic1111 + a popular CivitAI checkpoint
- For pure quality without custom needs, frontier closed models (Midjourney / Imagen / gpt-image-1 / Firefly) easier
See also
- Midjourney 🟩 🟦
- Imagen (Google) 🟩 🟦
- OpenAI Image (gpt-image-1) 🟩 🟦
- Adobe Firefly 🟩 🟦
- Flux (Black Forest Labs) 🟥
- Ideogram 🟥
- Hugging Face 🟩 🟦
- Replicate 🟩 🟦
- Together AI 🟩 🟦
- Stable Audio 🟥
- open-weights-vs-closed.md 🟥
- Multimodal (vision, audio) 🟩 🟦