视频后期，危！MiniMax H3手绘即特效，多模态的「Coding时刻」来了

TL;DR - MiniMax launched H3, its first open-weight video model, designed to generate publication-ready 2K videos with editing, typography, effects, music, and voice integrated end to end. Its multimodal inputs and private-deployment potential could make AI video more practical for commercial production.

• H3 accepts text, images, audio, and video as contextual inputs and supports targeted semantic editing.
• Its H3-Omni Transformer and customized captioning pipeline model relationships among multimodal references and target video.
• The model can generate animated text, transitions, visual effects, music, and emotionally synchronized dialogue within the output.
• The article reports H3 ranked first for video editing on Artificial Analysis and costs less per second than mainstream alternatives.