Stable Diffusion is a text-to-image diffusion model developed by Stability AI (UK). Since its first public release in August 2022, its defining characteristic has been openly published weights (the model’s trained parameter data). Under the Stability AI Community License, individuals and organizations with under $1 million in annual revenue can use it free of charge, including commercially. Because you can download the model onto your own PC or server and run it yourself — “self-hosting” — you can generate without worrying about image count limits or uploading your images to the cloud. That is the fundamental difference from cloud-only image generation services. The current generation, Stable Diffusion 3.5, ships in several variants — Large, Large Turbo, and Medium — and is also available through APIs and cloud services. Beyond images, Stability AI also offers generation models for video, audio, and 3D content.
Key Features
- Self-hosting via open weights: The model weights are published on Hugging Face and elsewhere, so you can run them locally on a PC or server with a supported GPU. You can build a setup where generated images never leave your machine
- Free commercial use under the Community License: Individuals and organizations under $1 million in annual revenue get free use for research, non-commercial, and commercial purposes alike. Organizations above that revenue threshold need a separate Enterprise License
- Current generation: Stable Diffusion 3.5: Built on MMDiT-X (an improved multimodal diffusion transformer), with better prompt adherence and more accurate human anatomy. Choose Large, Large Turbo, or Medium depending on your use case and available compute
- Available via API and cloud too: Without self-hosting, you can use it through Stability AI’s Developer Platform (credit-based API), AWS Bedrock, NVIDIA NIM, and other cloud offerings
- A rich ecosystem: Community-built tools like ComfyUI and AUTOMATIC1111 WebUI, additional training via LoRA, composition control via ControlNet — the kind of extensibility only an open model enables
- Models beyond images: Stability AI also publishes models for video (the Stable Video Diffusion family), audio (Stable Audio), and 3D generation, all under the same licensing framework
Pricing
| Usage model | Price | Key characteristics |
|---|---|---|
| Self-hosting (Community License) | Free | Commercial use free if under $1M annual revenue. Runs on your local GPU, no limit on image count |
| API (Developer Platform) | Credit-based (1 credit = $0.01, prepaid) | Credit consumption varies by model and feature. Generate via API with no GPU required |
| Enterprise License | Contact sales | For organizations with $1M+ annual revenue. Includes support and custom terms |
Note that stability.ai/pricing currently redirects to the plan lineup for Brand Studio, a team-oriented product (Trial free / Creator $19 per month / Core $50 per month / Enterprise contact sales). Brand Studio is a separate product from the developer-facing usage models in the table above, so if you want to run the model yourself or call the API, look at the developer portal (platform.stability.ai) instead. We were unable to obtain primary-source confirmation of the API credit rate as of August 2026, so please verify on the official site before you commit to any spend.
Pricing is current as of August 21, 2026. For the latest pricing, check the official site and the developer platform.
Pros and Cons
✅ Pros
- With self-hosting there is no cap on how many images you generate, and running costs are limited to electricity and hardware
- You can build a setup that never sends images to an external server, which makes sensitive material easier to work with
- A deep ecosystem — LoRA, ControlNet, ComfyUI — lets you train styles and control composition in fine detail
- A clear, permissive license: free commercial use under $1 million in annual revenue
- Even if running locally is impractical, you can use the same models through the API or cloud services
⚠️ Cons
- Local execution requires a machine with a suitably capable GPU (VRAM), and there is a learning curve to setting up the environment
- There is no official polished chat UI you can sign up for and use immediately, the way you can with ChatGPT or Midjourney — you have to pick your own tooling
- Getting exactly the image you want from a prompt alone takes knowledge of models and settings, plus trial and error
- Organizations with $1M+ in annual revenue must sign an Enterprise License
Comparison with Similar Services
| Criterion | Stable Diffusion | Midjourney | DALL·E (ChatGPT) | FLUX |
|---|---|---|---|---|
| Provider | Stability AI | Midjourney | OpenAI | Black Forest Labs |
| Delivery model | Open weights, self-host / API | Cloud-only (Web/Discord) | Cloud-only (ChatGPT/API) | Some open-weight models / API |
| Local execution | Yes | No | No | Yes, for some models |
| Free usage | Free when self-hosted (revenue threshold applies) | None (paid subscription) | Limited within ChatGPT’s free tier | Some free developer models |
| Extensibility | Extremely high — LoRA, ControlNet, etc. | Low | Low | High |
Who It’s For
- Individual creators and developers who want to generate large volumes of images without worrying about counts
- Organizations that need to generate images from sensitive material without sending it to an external cloud
- Anyone who wants fine-grained control over the generation process — style training with LoRA, composition control with ControlNet
- Developers who want to embed image generation into their own product and need a model with clear license terms
- People without a GPU on hand who still want to use an open model via API
Summary
Stable Diffusion is the definitive open model that brought open weights and self-hosting to image generation. There is a learning curve in setting up an environment and tuning prompts, but the freedom in generation volume, keeping your data local, and the depth of the ecosystem are strengths no cloud-only service offers. Start by trying Stable Diffusion 3.5 through the API or a tool like ComfyUI, then check your local environment and license terms (that $1 million annual revenue threshold) when you move toward serious production use.