Negative Prompt

Negative prompting is a technique used to improve the performance of generative models by specifying what should *not* be included in the output, complementing the usual positive prompts. Current research focuses on applying negative prompts across various modalities, including image generation (using diffusion models and algorithms like Perp-Neg), language models (improving instruction following and mitigating unwanted outputs), and robotic grasping (guiding object selection in complex scenes). This approach enhances model control and addresses issues like unwanted biases, improving the quality and safety of generated content and enabling more effective human-computer interaction in diverse applications.