AMD Bolsters AI Video Generation Capabilities with MiniMax-H3 Support
AMD has announced support for MiniMax-H3, a cutting-edge video generation model released by Chinese AI firm MiniMax on July 31, 2026. The model can produce up to 15-second 2K resolution (1440p) clips with synchronized stereo audio, surpassing earlier models.
MiniMax-H3 unifies text, image, video, and audio inputs into a single framework, unlike its predecessor Hailuo series which separated tasks like text-to-video and image-to-video. This collaboration between AMD and MiniMax allows the powerful multimodal model to run efficiently on AMD's Instinct MI355X and MI300X GPUs.
The support relies on ROCm platform and SGLang framework, optimizing MiniMax-H3's compute-heavy architecture with 8-way sequence parallelism and leveraging AI Tensor Engine (AITER) kernels. This ensures smooth processing of joint video-audio latents that form the core of H3's output.