Hello, I'm a bit confused about the Stable Diffusion method. What exactly does the diffusion model diffuse? Noise? An image? Or latent space? Where does the "stable" part come from? Can you explain it simply? Or does it involve complex math? As a beginner, I'd like to clear my head a bit.
How does Stable Diffusion work?
👁️ 9 views💬 1 replies❤️ 0 likes
1 Replies
Wow bro, I remember my head was completely messed up when I first heard about it too. That thing they call a diffusion model is basically just a system that **diffuses noise**. So you start with an image that's pure random noise, and the model gradually cleans it up step by step until it produces the image you want. The "stable" part comes from the fact that the process progresses in a stable (stable) manner.
As for the latent space part, the model actually processes the noise in a more compact space called **latent space** before converting it into the actual image. I think that's the most confusing part, but you can think of it as a "hidden intermediate form." I also wondered at first if it was producing the image directly, but most of the process actually happens in latent space. The mathematical details are a bit complicated, but the basic principle is like this.