MetaGenAI launches Fairy, a new video editing tool, which has attracted much attention for its amazing speed and powerful functions. Fairy supports natural language editing, enabling image style transformation (such as Van Gogh or Picasso style) and object or character transformation (such as converting a figure into a wooden sculpture or a metal knight sculpture). It uses an anchor-based cross-frame attention mechanism to ensure the temporal coherence and high fidelity of edited videos, and significantly improves the processing speed through diffusion models and parallel computing. It only takes 14 seconds to generate 120 frames of 512 × 384 videos. This has undoubtedly brought revolutionary changes to the field of video editing, making the realization of creativity more convenient and efficient.
MetaGenAI recently released Fairy video editing tool, which supports natural language editing and realizes style conversion and object role transformation. This tool is extremely fast, generating a 512×384 video at 120 frames in just 14 seconds. Key features include image style conversion (such as Van Gogh or Picasso style) and object or character conversion (such as a figure into a wood sculpture or a metal knight sculpture). Fairy uses an anchor-based cross-frame attention mechanism to ensure temporal coherence and high fidelity when editing videos. The diffusion model method is used to generate or edit video frames, and parallel computing is supported to significantly increase the processing speed. For detailed information, please refer to [Project Address](https://fairy-video2video.github.io/) and [Paper Address](https://arxiv.org/pdf/2312.13834.pdf).Fairy video editing tool brings new possibilities to the video editing industry with its efficient processing speed and powerful functions. In the future, we can look forward to Fairy's application in more fields and further performance improvements, bringing users a more convenient and powerful video editing experience. Interested users can visit the project address and paper address for more information.