huangserva/ComfyUI_MiniMaxH3_Director

What it solves

This project provides a set of pre-configured ComfyUI workflows for the MiniMax H3 Director model, simplifying the process of generating and editing high-quality video and audio. It enables users to perform complex video generation tasks—such as text-to-video, image-to-video, and video-to-video editing—without having to manually build the complex node graphs from scratch.

How it works

The project provides five distinct JSON workflow files that can be imported directly into ComfyUI. These workflows utilize the MiniMaxH3Director custom node and specific MiniMax H3 weights (such as fl2va and ref2va) to handle different generation modes:

• T2V: Text-to-video and audio generation.
• FL2V/I2V: First-and-last frame generation or image-to-video.
• R2V: Generation based on image, video, or audio references.
• V2V: Segment-by-segment editing based on a source video timeline.
• RV2V: Source video combined with reference images or audio, specifically useful for character replacement in videos.

Who it’s for

Content creators and AI video artists using ComfyUI who want to leverage the MiniMax H3 model for professional-grade video synthesis and editing, including character consistency and timeline-based editing.

Highlights

• Comprehensive Workflow Set: Covers T2V, I2V, V2V, and reference-based generation.
• Timeline-based Editing: Supports segment-by-segment editing and caching, allowing users to modify specific parts of a video without regenerating the entire sequence.
• ‘Character Replacement’ Capability: The RV2V workflow allows for video-to-video translation with reference images to change people in a video.
• Optimized Performance: Compatible with SageAttention for potential speed improvements.