AI-generated artwork is extremely popular right now. It is now possible to create photorealistic images directly on your PC without the need for external services such as Midjourney or DALL-E 2.
Stability AI is a tech startup that is working on the “Stable Diffusion” AI model, which is a complex algorithm trained on internet images. The company has officially released the Stable Diffusion model, which can be used to generate images from text prompts, after making a test version available to researchers. Unlike Midjourney and other models/generators, Stable Diffusion’s primary goal is to create photorealistic images, which has already sparked controversy over “deepfake” content. It can, however, be configured to mimic the style of a specific artist.
Stable Diffusion is unique in that it can generate images using a standard graphics card rather than remote (and expensive) servers. Stability AI currently recommends NVIDIA graphics cards, but full support for AMD and Apple Silicon is in the works.
Stable Diffusion has a ‘Safety Classifier’ mode that attempts to prevent the generation of offensive images, but because the model is open-source, it can be disabled when running on a PC. To avoid creating images that could deceive or harm others, web-based generators frequently prohibit users from using prompts that mention specific words or phrases. For better or worse, Stable Diffusion can generate more image types than most other services.
The Stable Diffusion model is intended to be used in conjunction with a generator, which serves as the actual interface for typing prompts and changing other options. One generator that can run locally on a computer is the Stable Diffusion Dream Script, which can be accessed via a command-line interface or a local web server. There’s also a plugin in the works for generating images from within Photoshop, and web-based versions are already available. For a step-by-step guide to running Stable Diffusion locally, see our guide.
Given Stable Diffusion’s open-source licencing model and impressive generation capabilities, it’s likely that most AI generators will adopt the new model.
