FreeNoise is a long video diffusion technique that requires no fine-tuning and achieves generation through noise rescheduling. It supports generating high-quality long videos and can be generated based on multiple text prompts. Compared to existing video generation models, FreeNoise has higher generation capabilities and time efficiency. It uses a window function to perform temporal attention adjustment on a series of noises to maintain content consistency. In addition, it has designed a novel motion injection method that supports video generation based on multiple text prompts. Compared to previous best methods, FreeNoise's time cost is almost negligible.