Skip to main content
The Mahiro node modifies the guidance function to focus more on the direction of the positive prompt rather than the difference between positive and negative prompts. It creates a patched model that applies a custom guidance scaling approach using cosine similarity between normalized conditional and unconditional denoised outputs. This experimental node helps steer the generation more strongly toward the positive prompt’s intended direction.

Inputs

Outputs

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 56f13cd14ffde21ab980850b62e07bdd4db6733a545f9329bf908c6030147981