Stable Diffusion: A Beginner’s Guide to Features, Pricing, License and Prompts
Stable Diffusion is a family of open image-generation models from Stability AI that turns text prompts into pictures, and you can run it on your own computer or through hosted apps. Because the model weights are downloadable, it has become one of the most flexible ways to make AI images without being tied to one website.
In this guide you will learn what it is, its main features, how to get started, how the license and pricing work, its pros and cons, good alternatives, and practical prompt tips with examples.

What is Stable Diffusion?
It is developed by Stability AI, a company that builds generative models for images, video, audio and 3D. The name covers several model generations. Older versions such as SD 1.5 and SDXL are still widely used by the community. The current flagship family is Stable Diffusion 3.5, which Stability AI released in October 2024 in three versions: Large (8.1 billion parameters), Large Turbo (a faster, distilled version that needs only a few steps) and Medium (2.5 billion parameters, designed for consumer hardware).
You describe an image in words, and the model generates it. Common use cases include concept art, illustrations, product mockups, marketing visuals, social media graphics, game assets and editing existing photos. Stability AI also offers commercial image services through its API, such as Stable Image Core and Stable Image Ultra, alongside the open models.
Key features
- Open model weights: You can download the models from Hugging Face and run them locally, offline and without per-image fees.
- Text-to-image: Generate images from a written prompt, with control over size, steps, guidance scale and seed.
- Image-to-image and inpainting: Start from an existing picture, change part of it, or fill in a masked area.
- Negative prompts: Most interfaces let you list things you do not want in the image, such as blur or extra text.
- Fine-tuning and add-ons: The community shares custom checkpoints, LoRAs and control tools that add styles, characters or pose control.
- Better text and prompt understanding in 3.5: Stability AI says SD 3.5 improves typography and handling of complex prompts compared with earlier versions.
- Many ways to run it: Local apps like ComfyUI, the Stability AI API, and third-party hosts such as Replicate, Fireworks AI and DeepInfra.
How to get started with Stable Diffusion
There are two main paths: run it locally for free, or use a hosted service. Here is the local route, which is the most common for beginners who have a suitable graphics card.
- Check your hardware. A dedicated GPU helps a lot. Stability AI states that SD 3.5 Medium needs about 9.9 GB of VRAM (excluding text encoders). Older models such as SD 1.5 run on much less memory.
- Install an interface. ComfyUI is a popular free option with a desktop app for Windows and macOS and support for NVIDIA, AMD, Intel and Apple Silicon GPUs. Stable Diffusion WebUI Forge and AUTOMATIC1111 are other well-known interfaces.
- Download a model. Create a free Hugging Face account, open the Stability AI model page, accept the license agreement and download the weights.
- Place the model file in your interface’s models folder and load it.
- Write a prompt, set the image size and steps, and click generate. For SD 3.5 Large, the model card suggests around 28 steps and a guidance scale of 3.5 to 4.5.
- Adjust and repeat. Change the prompt, seed or settings until you get the result you want.
If you do not have a strong GPU, sign up for the Stability AI developer platform or a third-party host and generate images in the cloud instead.
Stable Diffusion pricing and free plan
Running the models locally is free in the sense that you pay nothing per image. Your costs are your own hardware and electricity. Licensing depends on the model version:
- SD 3.5 models: Covered by the Stability AI Community License. As of September 2026, it allows free use for research, non-commercial and commercial purposes for individuals and organizations with less than USD 1 million in annual revenue.
- Larger businesses: Organizations above that revenue threshold that use the models commercially need an Enterprise License, with custom pricing from Stability AI.
- Older models: SD 1.5 and SDXL were released under different (CreativeML OpenRAIL-style) licenses, so read the license on each model page before commercial use.
Hosted options cost money. The Stability AI API uses a credit system, where each model or service uses a different number of credits per image. Third-party hosts and apps set their own prices. Because rates change, check the official Stability AI pricing page for the latest numbers before you commit.
Pros and cons
| Pros | Cons |
|---|---|
| Free to run locally with no per-image fees | Local setup takes time and some technical comfort |
| Full control over settings, seeds and workflows | Needs a capable GPU for newer models |
| Huge community of models, LoRAs and tutorials | License terms differ between model versions |
| Works offline, so prompts and images stay on your machine | Results often need more prompt tuning than simple chat-based tools |
| Available through APIs and many hosted apps | Community add-ons vary in quality and safety |
Who should use Stable Diffusion?
It suits people who want control and flexibility. It is a strong fit for:
- Hobbyists and artists who enjoy experimenting with styles and custom models.
- Developers who want to build image features into apps via local models or the API.
- Small businesses and creators under the license revenue threshold who need many images without per-image costs.
- Privacy-minded users who prefer to keep their work on their own computer.
If you just want a quick image with no setup, a browser-based generator may be easier to start with.
Stable Diffusion alternatives
- Midjourney: A popular hosted generator known for its polished, artistic look. It is a paid service with no local option.
- ChatGPT image generation (OpenAI): Create images by chatting in plain language. Usage limits depend on your plan.
- Adobe Firefly: Adobe’s image model, built into Photoshop and other Creative Cloud apps, aimed at commercial design work.
- FLUX by Black Forest Labs: Another family of image models with open-weight versions that also run in ComfyUI. Check each version’s license.
Whichever tool you choose, good prompts matter. If you want help writing them, our free Image Prompt Builder lets you combine subject, style, lighting and camera settings into a ready-to-use prompt. You can also browse our prompt library for ready-made starting points to adapt.
Tips for better results
- Be specific: Name the subject, setting, style, lighting and camera angle instead of a single vague phrase.
- Put key details first: The most important parts of the image should come early in the prompt.
- Fix the seed when refining: Keep the same seed and change one word at a time to see what each change does.
- Use the recommended settings: Each model has suggested steps and guidance values on its model card. Very high guidance can make images look harsh.
- Generate at native sizes: Create images near the model’s intended resolution, then adjust them afterwards.
Example prompts
- A cozy coffee shop on a rainy evening, warm light through the windows, people reading at wooden tables, cinematic photo, 35mm lens, shallow depth of field
- Minimalist flat illustration of a mountain lake at sunrise, pastel colors, clean shapes, soft gradients, poster design
- Product photo of a glass perfume bottle on a white marble surface, soft studio lighting, subtle reflections, high detail, centered composition
Negative prompt example
blurry, low quality, distorted hands, extra fingers, watermark, text, logo, oversaturated, cropped
Negative prompts tend to have the strongest effect on older models like SD 1.5 and SDXL. With SD 3.5, a clear positive prompt usually matters more. When your image is ready, use our free Resize Image tool to fit it to social media or website dimensions, and convert the image to JPG, PNG or WebP if you need a different format.
Frequently asked questions
Is Stable Diffusion free?
The model weights are free to download and run locally. SD 3.5 is free under the Stability AI Community License for individuals and organizations with under USD 1 million in annual revenue. Hosted services and the Stability AI API charge for usage.
Can I use the images commercially?
Often yes, but it depends on the model version and your organization’s revenue. Read the license on the model page, and consider an Enterprise License if your business is above the Community License threshold.
Do I need a powerful computer?
For newer models, a GPU with around 10 GB of VRAM or more makes things much smoother. Older models run on less. If your computer is not powerful enough, use a hosted service instead.
Which version should a beginner use?
SD 3.5 Medium is a good balance of quality and hardware needs. SDXL is also a solid choice because of its large library of community models and tutorials.
Sources and further reading: Introducing Stable Diffusion 3.5 (Stability AI), Stability AI License, Stable Diffusion 3.5 Large on Hugging Face.
More AI tool guides
Free tools for this
Everything runs in your browser. Your files never leave your device.