Stable Diffusion
: How to use it, features, and the business problems it solves
Add bookmark
What is Stable Diffusion?
A family of open-weight image-generation AI models developed by Stability AI. The current flagship is Stable Diffusion 3.5 (Large / Large Turbo / Medium). You can download the weights for free and run them locally (GPU required), and it is also usable via the official API, DreamStudio, ComfyUI, and more. Under the Community License, commercial use is free for under $1M in annual revenue (prompts are mainly in English).
Business problems it solves
About "Stable Diffusion"
What is Stable Diffusion
Stable Diffusion is a family of image-generation AI models developed by Stability AI and released with open weights. It generates images from text and is not a single product but a collective name for multiple versions and derivative models. Its biggest feature is that you can download the model weights for free and run them in your own environment. Unlike closed image-generation AIs, it allows local execution, free fine-tuning, and offline operation, and it has a huge community and ecosystem including LoRA and ControlNet.
The current flagship, Stable Diffusion 3.5, is centered on English prompts. The claim that "the latest version supports Japanese input" is not accurate; the Japanese-specialized one is a separate model, "Japanese Stable Diffusion XL" (released 2023). Also note the license is not pure open source but open-weight (Stability AI Community License) (as of August 2026).
An image-generation AI whose weights you can run yourself
The essential strength of Stable Diffusion is that you can obtain the model itself and use it freely. Whereas closed services are limited to use via an API or app, Stable Diffusion lets you download the weights, run them on your own GPU, fine-tune with your own data, and operate offline or on your own infrastructure. This freedom has fueled an ecosystem of countless custom models and extensions (LoRA, ControlNet).
Closed image-generation AI
- Use only via API/app
- The model can't be modified or self-hosted
- Data is sent to the service
Stable Diffusion
- Download weights and run locally
- Freely tune with fine-tuning and LoRA
- Offline or self-hosted operation
Current models
01SD 3.5 Large
An 8.1-billion-parameter top-quality model. Excellent prompt adherence, up to 1 megapixel.
02SD 3.5 Large Turbo
A distilled version of Large that generates fast in 4 steps.
03SD 3.5 Medium
2.5 billion parameters. A lightweight version that runs even on consumer GPUs with about 10GB of VRAM.
Besides these, the previous-generation SDXL 1.0 / SDXL Turbo are also still widely used.
How to use / delivery forms
01Run locally
- Download weights — get them from Hugging Face and GitHub (GPU required)
- UI — operate with ComfyUI, Automatic1111, Fooocus, etc.
02Official API and services
- Stability AI Platform — API (Stable Image Ultra / Core)
- DreamStudio — the official web UI
03Via partners
- Hosting — also usable from services such as Replicate
Pricing
Downloading the model weights themselves is free (you need a GPU to run them). The official API is credit-based, with 1 credit costing roughly $0.01. For reference, Stable Image Ultra (equivalent to SD 3.5 Large) is about $0.08/image, Core about $0.03/image, and SD 3.5 Medium about $0.035/image. API Membership is roughly $20/month for 6,000 credits.
The license is the Stability AI Community License: non-commercial use is free, and commercial use is also free if the individual's or organization's annual revenue is under $1 million. Commercial use at $1 million or more in annual revenue requires a separate Enterprise license. Ownership of generated images belongs to the user. Amounts and terms may change, so check the official site for the latest (as of August 2026).
How it differs from other image-generation AIs
Image-generation AIs split into "closed service type" and "open type that distributes weights." Stable Diffusion is the leading example of the latter, standing out for customizability and the breadth of its ecosystem.
| Aspect | Stable Diffusion | Midjourney | FLUX |
|---|---|---|---|
| Delivery | Open weights (weights distributed) | Closed (service only) | Open weights |
| Execution | Local, API, various UIs | Web / Discord | Local, API |
| Customization | Rich fine-tuning and LoRA | Limited | Expanding |
If you choose by style or ease of use, Midjourney; for an integrated creative environment, Leonardo.Ai; for a newer open model, FLUX are comparison options. When you want to run a model in your own environment and customize it freely, Stable Diffusion is a strong choice.
The information on this page is based on content published on the official website and model cards as of August 2026. Model generations, specifications, and pricing may change, so check the official website for the latest.
