MiniMax h3

Omnidirectional Input Support

Upload up to 9 reference images, up to 3 reference videos (each 2–15 seconds, totaling no more than 15 seconds), and up to 3 reference audio clips. Mixed inputs can be combined into up to 12 files in total, expanding creative possibilities. With MiniMax H3, you can upload up to 50MB per video, 30MB per image, and 15MB per audio file, with a total capacity of up to 64MB per request.

Powerful @ Reference System

Specify how each asset is used with just natural language @ mentions. Reference @Image1 for character style, @Video1 for camera movement, @Audio1 for the soundtrack, and more, giving you complete creative control. Prompts can be up to 7,000 characters long, and images can be in JPG, PNG, WEBP, HEIC, or HEIF formats, while videos can be in H.264 or H.265 formats.

Key Strengths

According to MiniMax's own benchmarks, the standout strengths of MiniMax H3 are precise instruction understanding, rendering of on-screen text and brand information, and motion transfer based on video references. MiniMax states that multi-shot composition was explicitly modeled from the pre-training stage. Based on MiniMax's own metrics, the per-second cost for 2K resolution is less than one-third that of major models.