Stable Diffusion
Stable Diffusion is the open source text-to-image model family that put AI image generation in the hands of everyone with a decent GPU. Released in 2022, it kicked off an entire ecosystem of local generation tools, community checkpoints and fine-tuned models that you can run free, offline and private.
Quick Facts
| Developer | Stability AI |
|---|---|
| First released | 2022 |
| Latest version | Stable Diffusion 3.5 (2024); Stable Diffusion 4 (2025) |
| Language | Python (PyTorch) |
| License | Open source and open weights variants |
| Pricing | Free to download and run; needs a capable GPU |
| Platform | Runs locally, in the cloud and via hosted services |
What is Stable Diffusion?
Stable Diffusion is a family of open source text-to-image models developed by Stability AI with academic partners including CompVis at LMU Munich. When the first public version landed in August 2022, it was a breakthrough: for the first time, anyone with a decent graphics card could generate images from text on their own machine, with no cloud service and no content filters beyond their own choices.
The model uses a technique called latent diffusion, which compresses images into a smaller latent space before the diffusion process, making generation fast enough to run on consumer hardware. That technical choice is why the local AI art scene exploded around it.
Since then the family has grown through many versions, including SDXL for higher resolution and Stable Diffusion 3.5 and 4 for better prompt understanding, plus a huge community ecosystem of fine-tuned checkpoints, LoRA adapters and tools like ControlNet.
Key Features
- Text-to-image and image-to-image generation from a single prompt
- Open weights you can download and run locally, free and offline
- Fine-tuning and LoRA support for training custom styles and subjects
- ControlNet and similar tools for precise control over composition
- Inpainting and outpainting for editing existing images
- A massive community ecosystem of checkpoints and plugins
- Runs on consumer GPUs with tools like ComfyUI and AUTOMATIC1111
How to get started
The easiest way to try Stable Diffusion is a hosted service, many of which offer free tiers, or a web tool like DreamStudio from Stability AI itself. You type a prompt, pick a style and download the result, no installation needed.
To run it locally, install a frontend like AUTOMATIC1111 or ComfyUI, download a model checkpoint from Hugging Face or Civitai, and start generating. A GPU with 8 GB or more of VRAM is the sweet spot, though smaller models run on less.
- Try a hosted service or DreamStudio for a zero-install taste
- Install AUTOMATIC1111 or ComfyUI on your machine
- Download a checkpoint from Hugging Face or Civitai
- Write a prompt and generate your first image
- Experiment with LoRAs and ControlNet for more control
- Join the community on Civitai and Discord for inspiration
Use cases
- Personal art and experimentation without paying per image
- Concept art and asset exploration for indie games
- Training custom models and LoRAs on your own art style
- Private or sensitive projects where cloud tools are not an option
- Research and education in generative AI
- Building products: many apps and services run on Stable Diffusion under the hood
Pricing and licensing
The models themselves are free to download and run. Early versions used the CreativeML Open RAIL-M license, which allows commercial use with some restrictions, while newer releases such as Stable Diffusion 3.5 use Stability AI's own community license with a revenue threshold for very large companies. Stable Diffusion 4 introduced a similar open models license.
Running locally costs you electricity and hardware, not tokens. Hosted services and APIs are priced by the image. If you want to sell products built on newer versions, read the license terms for your version first, since the details differ between releases.
Pros and cons
Freedom is the whole point. You own your workflow, your images never leave your machine, and the ecosystem of checkpoints and tools is unmatched by any closed service. Cost is a big win too, since local generation is effectively free once you have the hardware.
The downsides are setup friction and consistency. Getting a good local install takes an afternoon, results vary a lot between checkpoints and prompts, and out of the box it is less polished than commercial tools like Midjourney. You trade convenience for control.
Alternatives
- Midjourney: the most artistic commercial generator, subscription only
- DALL·E 3: OpenAI's model, integrated into ChatGPT
- FLUX: a newer open weights family from Black Forest Labs
- Adobe Firefly: commercially safe generation in Adobe's tools
- Ideogram: excellent for text rendering in images