Modular Pipelines

Modular diffusers is a new approach to building and using diffusion pipelines that allows for greater flexibility and customization
It is designed to be compatible with the existing Hugging Face Diffusers library, while also providing additional features and capabilities
See https://huggingface.co/docs/diffusers/modular_diffusers/overview for more information

SD.Next implements modular pipelines in a way that allows for seamless integration with existing workflows

So why mention modular pipelines?
Because some of the new features are only available in modular pipelines, and some of the existing features are only available in standard pipelines

Some pipelines are modular-only, some are standard-only and some are available in both formats
Examples: - Standard-only: Krea-2, Anima - Modular-only: MiniMax - Hybrid: Z-Image

Hybrid pipelines

If selected in settings -> model loading -> model convert to modular,
SD.Next will automatically attempt to convert standard pipelines to modular pipelines when loading models, if the model supports it. If the model does not support it, SD.Next will fall back to standard pipelines automatically.

Modular-only features

Guidance methods

While standard pipelines support several guidance methods, modular pipelines support additional guidance methods that are not available in standard pipelines