MiniMax Releases MiniMax H3: An Omni-Modal Video Model That Generates 15-Second 2K Clips With Native Stereo Audio

That Generates 15-Second 2K Clips With Native Stereo Audio">

MiniMax releases MiniMax H3 , a general-purpose multimodal generation model. MiniMax H3 is not a text-to-video model with add-ons. MiniMax describes it as a general-purpose multimodal generation model that reads text, images, video, and audio as one unified context and returns video with native stereo sound. The mains specs include: 2K output, 4–15 seconds, integer durations only .

Previous video stacks split into text-to-video, image-to-video, first-and-last-frame, subject reference, motion reference, and video editing, each often a separate expert model. MiniMax H3 folds those into one pretraining paradigm where reference and editing relationships are expressed in natural language. MiniMax’s example prompt makes the point: reference the camera movement from Video 1, have the character in Image 2 sing, match the vocals to Audio 3.

Today: yes, through the API and no, on your own hardware. MiniMax launched H3 on July 31, 2026 with the model live in the platform API under the model ID MiniMax-H3 and in the consumer Hailuo AI app.

Industries : MiniMax positions MiniMax H3 for advertising, branding, e-commerce, product design, UI/UX, and gaming along with film pre-visualization and retail catalog media.

Applications : Ad variant generation, product and listing videos, animated posters, film title sequences, website hero loops, character-consistent game cinematics, and video-to-video motion transfer.

The video generation guide documents three entry modes: text-to-video, first/last-frame image-to-video, and reference generation. Behind one endpoint and an asynchronous three-step flow: create a task, poll task_id , download content.url .

Contextual Omni Representation : MiniMax rebuilt captioning so it describes the relationship between context and target video, not just the target. Most source material requires roughly 100K tokens of inference, distilled to about 4K tokens on average. Language is the bridge that turns a fixed task set into an open, descriptive one.

H3-VAE : A full tokenizer overhaul. Its high compression ratio delivers a stated 4× gain in effective sequence length , cutting training and inference cost and it is the enabling technology for native 2K.

H3-Omni Transformer : MiniMax explicitly set aside the Hailuo-02 architecture here. Multimodal context tripled sequence-length variance, so the training architecture separates understanding and generation workloads and tunes hardware utilization for each. Reported result: end-to-end training throughput up nearly 30% .