The NVIDIA Sol team released an open-source acceleration method for MiniMax-H3 video generation that runs faster than playback. It is available via the Reactor API so users can test it before committing GPU resources.
A new arXiv paper presents Sol-H3, a full-stack inference pipeline accelerating the 33-billion-parameter MiniMax-H3 open-source video diffusion model on both cloud (NVIDIA GB200) and edge (DGX Spark) hardware. It combines a two-stage cross-resolution generation scheduler with a Recursive Self-Improvement kernel-search loop, achieving up to 30x end-to-end speedup and 20% lower memory.
A new arXiv paper presents Sol-H3, a full-stack inference pipeline accelerating the 33-billion-parameter MiniMax-H3 open-source video diffusion model on both cloud (NVIDIA GB200) and edge (DGX Spark) hardware. It combines a two-stage cross-resolution generation scheduler with a Recursive Self-Improvement kernel-search loop, achieving up to 30x end-to-end speedup and 20% lower memory.
NVIDIA's SANA team introduced Sol Engine, a technique that splits MiniMax H3 video generation into a 4-step low-res draft and a 3-step LTX refinement pass with Sol-Attn. This reduces 768p latency on a single GB200 from 414 seconds to 14.93 seconds, a 27x speedup.
NVIDIA's SANA team introduced Sol Engine, a technique that splits MiniMax H3 video generation into a 4-step low-res draft and a 3-step LTX refinement pass with Sol-Attn. This reduces 768p latency on a single GB200 from 414 seconds to 14.93 seconds, a 27x speedup.
Community demo shows Nunchux generating 5-second MiniMax-H3 videos in 1.3s on AMD MI355X
Community demo shows Nunchux generating 5-second MiniMax-H3 videos in 1.3s on AMD MI355X
MiniMax released H3, a unified multimodal model that generates 5–15 second video clips at native 2K resolution with stereo audio. SGLang Diffusion provides day-0 serving support, enabling local inference on consumer/workstation GPUs like 2× RTX 5090 or 1× RTX Pro 6000.
MiniMax released H3, a unified multimodal model that generates 5–15 second video clips at native 2K resolution with stereo audio. SGLang Diffusion provides day-0 serving support, enabling local inference on consumer/workstation GPUs like 2× RTX 5090 or 1× RTX Pro 6000.
An open-source real-time video generation release from MiniMax, built on vLLM-Omni and Hao AI Lab's FastH3 with NVIDIA hardware support. Real-time generation enables interactive video applications, and the open baseline lets the community improve it. It targets developers and researchers wanting a public, extensible foundation for streaming video generation.
An open-source real-time video generation release from MiniMax, built on vLLM-Omni and Hao AI Lab's FastH3 with NVIDIA hardware support. Real-time generation enables interactive video applications, and the open baseline lets the community improve it. It targets developers and researchers wanting a public, extensible foundation for streaming video generation.