🛰️ Daily AI Frontier
‹ back to 2026-08-01

比 Seedance 便宜七成的模型,宣布开源:视频生成行业变天了?

雷峰网 (AI科技评论) Multimodal & Generative 2026-07-31
Representative image for 比 Seedance 便宜七成的模型,宣布开源:视频生成行业变天了?

TL;DR - MiniMax launched H3, a flagship multimodal video model priced at one-third of Seedance 2.0, with plans to open-source it. Its integrated generation and editing capabilities could provide a customizable foundation for commercial video workflows.

  • H3 jointly handles text, images, audio, and video for generation, editing, synchronization, typography, and post-production.
  • Its H3-VAE reportedly yields 4× sequence-length gains, while H3-Omni Transformer improves end-to-end training throughput by nearly 30%.
  • The model supports native 2K output through context-aware regeneration rather than a separate super-resolution module.
  • Open weights could enable private deployment, industry-specific fine-tuning, and integration into enterprise creative toolchains.

view merged work →