The reporter learned that recently, the Efficient-Efficiency Model team composed of Nvidia and MIT researchers released the model weights for the video enhancement project OPEN-H3-REGENERATE-2K on HuggingFace. The supporting inference code has already been opened in the NVLabs/SANA warehouse. Based on MiniMaxH3, the solution regenerates low-resolution video drafts into high-resolution videos. It supports 2K output of 2560 x 1440, and can achieve native 4K regeneration of up to 3840 x 2160.

The reporter learned that recently, the Efficient-Efficiency Model team composed of Nvidia and MIT researchers released the model weights for the video enhancement project OPEN-H3-REGENERATE-2K on HuggingFace. The supporting inference code has already been opened in the NVLabs/SANA warehouse. Based on MiniMaxH3, the solution regenerates low-resolution video drafts into high-resolution videos. It supports 2K output of 2560 x 1440, and can achieve native 4K regeneration of up to 3840 x 2160.