Rimagination/h3lite

A hardware-aware Codex skill for local MiniMax H3 video generation through ComfyUI

What it solves

H3 Lite provides a way to locally deploy and run the MiniMax H3 video generation model on Windows machines with NVIDIA GPUs, specifically targeting users with low VRAM. It simplifies the complex process of setting up ComfyUI, managing model weights, and crafting high-quality video prompts for AI agents like Codex and WorkBuddy.

How it works

The project implements a "Skill" for AI agents that automates the model's deployment and execution. It uses two optimized component sets (Set A for low VRAM/fast paths and Set B for FP8 compatibility) to balance performance and quality. The workflow follows a four-step process: intent routing, reference image anchoring, prompt enhancement, and generation with verification. It supports multiple generation modes: Text-to-Video (T2VA), Image-to-Video (I2VA), First-and-Last-Frame-to-Video (FL2VA), and Reference-to-Video (Ref2VA).

Who it’s for

Windows users with NVIDIA GPUs who want to generate high-quality videos with native audio locally, and developers building AI agents that need a video generation capability.

Highlights

  • Low VRAM Optimization: Specifically tuned for GPUs like the RTX 4070 Laptop (8GB) and RTX 4060 Ti (16GB).
  • Agent-Ready: Designed as a skill for Codex and WorkBuddy, allowing users to trigger video generation via natural language.
  • Multimodal Input: Supports using images, videos, and audio as references to maintain character and scene consistency.
  • Integrated Monitoring: Includes a native Windows progress window that tracks sampling, VRAM usage, and estimated time of arrival (ETA).
  • Native Audio: Generates videos with synchronized native soundscapes and music.

Related

  • Dispatch
    MiniMax H3 Support in ComfyUI

    ComfyUI introduces day-zero support for MiniMax H3, an open-weights omni-modal video model capable of generating 2K video with native stereo audio on consumer hardware.

  • Project
    huangserva/ComfyUI_MiniMaxH3_Director

    A collection of ComfyUI workflows for MiniMax H3 Director, enabling text-to-video, image-to-video, and advanced video-to-video editing with reference materials.

  • Project
    LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler

    A custom ComfyUI node set for Minimax H3 that provides neural latent upscaling and tiled re-sampling to accelerate high-resolution video generation while avoiding VAE round-trip overhead.

  • Project
    nkxx188/ComfyUI-MiniMaxH3-Easy

    A compact ComfyUI interface for MiniMax H3 that simplifies text-to-video, image-to-video, and reference-based video generation with a unified media input system.

  • Project
    antirez/h3.c

    A native Metal inference engine for MiniMax-H3 on Apple Silicon, enabling local text-to-video and text-to-audio generation with advanced memory and performance optimizations.