Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Open-H3-Regenerate-2K Released for 2K and 4K Video Regeneration

AI summary

NVIDIA and MIT researchers on the Efficient-Large-Model team released the Open-H3-Regenerate-2K weights on Hugging Face and inference code in the NVlabs/Sana repository. Based on MiniMax H3, the deployable implementation regenerates low-resolution drafts at 2560×1440 and supports up to 3840×2160 native 4K output. The team reports that a prewarmed five-second 2K processing pass on one NVIDIA B200 GPU fell from 87.26 seconds to 22.45 seconds, while its UniPercept score reached 60.0 versus 53.4 for the official API.

In-depth insight

The release provides an accessible, locally deployable path for high-resolution video regeneration around the open-weight MiniMax H3 ecosystem, with reported gains in both processing speed and perceived image quality.

近日,由英伟达与麻省理工学院研究人员组成的 Efficient-Large-Model 团队,在 Hugging Face 发布视频增强项目 Open-H3-Regenerate-2K 的模型权重,配套推理代码已在 NVlabs/Sana 仓库开放。该方案基于 MiniMax H3,将低分辨率视频草稿重新生成为高分辨率视频,支持 2560×1440 的 2K 输出,最高可实现 3840×2160 的原生 4K 重生成。

MiniMax H3 的完整系统包含多模态上下文处理、基础音视频生成和高分辨率重生成等环节。其中,已开放权重的 H3-Base 负责生成 768P 音视频,官方 H3-Regenerate-2K 模块通过 API 提供服务。此次发布的项目复用 H3 基础模型的生成能力,为高分辨率重生成环节提供了一套可自行部署的开放实现。

在细化模型的训练阶段,研究团队保持 H3 Transformer 主干与视频 VAE 的原有参数不变,仅训练新增的细化 LoRA。推理时,细化 LoRA 与 LightX2V FL2VA Turbo、LMS Sharp 等社区适配器合并使用,通过三步去噪,在目标分辨率下重新生成画面细节。视频草稿中的场景、布局和运动信息,则为这一过程提供基础。

#NVIDIA#Massachusetts Institute of Technology#Efficient-Large-Model#MiniMax#Open-H3-Regenerate-2K#MiniMax H3#H3-Base#H3-Regenerate-2K#NVlabs/Sana#NVIDIA B200#LightX2V FL2VA Turbo#LMS Sharp

Original · 3 min readRead the original
Group chat

No comments yet. Start the conversation.