Unverified50% confidenceFactExact time
Minimax是字节跳动旗下的多模态大模型，H3代表其图像生成版本
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/7/2026
First Seen
Valid until: 12/6/2026
Sources
Related Entities
Related Claims
UnverifiedH3 Minimax是Minimax公司的视频生成引擎，采用了稀疏注意力（Sparse Attention）机制78% similarVerifiedMiniMax M3 is a multimodal model that supports text, video, images, and other modalities78% similarVerifiedMiniMax-H3具备图像到视频（Image-to-video）和基于参考的生成（reference-based generation）两项能力78% similarUnverifiedMiniMax H3模型采用4帧合一的时间压缩策略，一个潜在token跨越4帧画面77% similarUnverifiedMiniMax H3在图生视频任务中通过时序建模和身份保持模块，能完整保留原图中的人物形象与画风特征，不出现形象崩坏77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/873196API
curl https://kongchang.com/api/v1/knowledge/claims/873196MCP
get_claim(id=873196)