The article introduces a novel approach called FABRIC, which enhances the generation process of personalized diffusion models through iterative feedback. FABRIC leverages positive and negative feedback images to modulate the generation process, fine-tuning the generated images according to user preferences, thereby offering a more controlled and interactive text-to-image generation experience. Utilizing the self-attention module in U-Net, FABRIC injects additional information through reference images to create new images that resemble the reference images. Through multiple rounds of positive and negative feedback, FABRIC allows for iterative refinement of the generated images, delivering more accurate results.