FramePack
High-quality video generation, fast local processing, multi-modal input support.
- Creator
- 平台
- Web
FramePack is a revolutionary video diffusion technology designed specifically for consumer-grade GPUs with 6GB of video memory, enabling high-quality video generation. It uses frame context packing technology to improve video generation quality and prevents quality drift in long video sequences through bidirectional sampling technology. After Teacache optimization, the video generation speed is increased to approximately 1.5 seconds per frame. It supports local video generation without relying on cloud processing or expensive GPU rentals. It also supports multimodal inputs, including text prompts and image inputs. An open-source implementation is provided, supporting customization and community contributions.
