Hypernetworks are small auxiliary networks that “sit on top” of a Stable Diffusion model to modify its output behaviour without changing the base model’s weights.

How do they work?

• During image generation, hypernetworks adjust or bias certain internal features of the main model.
• They act like a style filter or modifier, influencing textures, colours, or subtle stylistic elements.
• They are usually much smaller than full models or LoRA files, often a few MBs.

Why use Hypernetworks?

• To add or switch styles quickly without retraining or loading big models.
• To experiment with subtle visual effects or texture changes.
• Useful for older Stable Diffusion versions (v1.4/1.5) where hypernets were more popular.

Limitations:

• Hypernetworks usually provide subtle changes, not dramatic new subjects or concepts.
• They can sometimes cause unpredictable results if combined improperly.