Open-H3-Regenerate-2K-IC is the latest open-source MiniMax-H3 high-resolution regeneration model from the NVIDIA/MIT Efficient-Large-Model team. It's specifically designed to upscale low-resolution drafts to higher resolutions while preserving as much of the original draft's key details as possible. Core Features: Function: Regenerates MiniMax-H3 768P drafts (typically 1344×768) to 1080P / 2K (2560×1440, default) / 4K (3840×2160). In-Context (IC) Highlight: The draft is completely preserved within the MiniMax-H3 Ref2VA transformer context as a reference video, and then a high-resolution version is generated using IC LoRA. Therefore, it better preserves: - Faces and identities - Lip synchronization - On-screen text/subtitles - Overall layout and composition - Compared to previous non-IC versions, there is less drift, and text and facial features are more stable. Suitable for scenarios where a 768P video has already been generated using MiniMax-H3 (or a compatible draft), and a higher resolution is desired while maintaining minimal distortion of faces, lip movements, and text. It is one of the open-source alternatives to the official H3-Regenerate-2K API. Hugging Face Weights: https://huggingface.co/Efficient-Large-Model/Open-H3-Regenerate-2K-IC ComfyUI Nodes/Workflow: https://(github.com)/slmonker/ComfyUI-OpenH3Regenerate2KIC/tree/main