MiniMax 发布 H3 视频模型，集成剪辑特效与多模态理解
MiniMax H3 将视频后期全流程集成进模型，支持多模态输入与商业级文字特效，显著降低视频制作门槛。
MiniMax 发布 H3 视频生成模型，将剪辑、字幕、转场、配乐、特效等后期能力端到端集成，支持图文音视频全模态输入。模型能理解文字与画面关系、生成稳定文字特效，并可通过音频驱动台词情绪。在 Artificial Analysis 榜单上视频编辑能力排第一，API 价格低于主流模型。
正文摘录
- title: Video Post-Production in Danger? MiniMax H3: Hand-Drawn Effects as Special FX, Multimodal's "Coding Moment" Has Arrived - sourcecompany: QbitAI - bodymarkdown: Over the past year, video models have advanced by leaps and bounds thanks to the Scaling dividend, but one thing hasn't changed: they only provide you with "footage." And after the footage is generated? You still have to drag it into Premiere Pro, add subtitles, grade colors, create transitions, add music, apply templates... a whole "ten thousand other tasks" waiting for you. But this model integrates editing logic, typography, transition design, rhythm control, background music, and visual effects—all "end-to-end" — directly into the model itself. On the freshly updated Artificial Analysis leaderboard, H3 also unsurprising…