Songssx/ComfyUI-MiniMaxH3-TimelineDirector

What it solves

This project provides a comprehensive timeline-based editing and generation interface for the MiniMax H3 video model within ComfyUI. It solves the problem of creating long-form, seamless videos by splitting generation into continuous segments that avoid the quality degradation, blur, and visible seams typically found in multi-segment video generation. It also simplifies the management of multimodal references (images, videos, and audio) for complex tasks like digital human lip-syncing and character replacement.

How it works

The plugin introduces a "Material Planner" that acts as a compact editing surface where users can arrange reference media on a timeline. It uses several key techniques to ensure continuity:

• Finite Segment Sampling: Generates video in overlapping segments, carrying the latent tail of the previous segment directly into the next to avoid RGB decode/re-encode cycles.
• Drift-Control and Soft AV: Employs adaptive masking and audio continuity techniques to ensure visual and audio transitions between segments are smooth.
• Locked Original Audio: Allows users to inject source waveforms directly into the AV-latent path, ensuring lip motion remains audio-driven while preserving the original recording.
• Two-Stage Sampling: Implements a progressive-resolution approach (e.g., 75% low-res and 25% high-res) to maintain high quality over long durations without progressive quality loss.

Who it’s for

Content creators and AI video artists using ComfyUI and the MiniMax H3 model who need to produce professional, long-form video content, digital humans, or complex video-to-video edits with precise temporal control.

Highlights

• Unlimited-length video generation: Creates long videos through continuous segmenting without needing duplicated sampler chains.
• Integrated Timeline: A multi-clip timeline with trimming, splitting, and snapping for managing reference media.
• Two-stage sampling: Built-in progressive-resolution sampling to prevent blur and seams in long videos.
• Digital Human support: Specialized modes for lip-sync and singing avatars with locked source audio.
• Omni Prompt Bridge: Integration with prompt rewriters to allow multimodal understanding of reference materials before generation.
• Preset System: Ability to export and import complete material and configuration presets for easy project sharing.