Diffusion model
Published Updated
What is Diffusion model?
A class of AI models that generate images, video, or audio by taking random digital noise and gradually refining it into a clear, structured output based on a prompt.
Diffusion models (like Stable Diffusion or Midjourney) are trained by taking clear images, systematically destroying them by adding static noise, and learning how to reverse the process. During generation, the model starts with pure static and iteratively denoises it, guided by text descriptions, until a coherent image emerges.
How it relates to Orbit
Sivants Studio, Orbit’s built-in image and video workspace, generates images with diffusion-based models through fal.ai and Recraft. See Sivants Studio for details.
When you would use it
You refer to diffusion models when discussing the underlying architecture powering modern generative visual and audio AI tools.
Common operations
- Not applicable.