Although a method to apply Negative Prompts to the FLUX.1 model by adapting Dynamic Thresholding has emerged, there is a phenomenon where applying Negative Prompt drastically reduces the image generation speed.

This method optimizes the speed of image generation with Negative Prompts by applying them only to the initial steps that determine the overall composition.

image 1: workflow

image 2: result image - empty 3 steps negative prompt (car is appeared)

image 3: result image - with 3 steps negative prompt 'car' (GOOD!)

image 4: result image - without negative prompt (car is appeared)

image 5: result image - with full steps negative prompt 'car' (blurred)