🎬 H3 Planner — Long-Form AI Music Video Pipeline

A complete custom node suite for creating multi-minute AI music videos with MiniMax H3 — even on 8GB–12GB VRAM. This is the exact system used to build full-length AI music videos from a single prompt or lyric sheet, without hitting memory limits or losing character consistency across shots.

Load, manage, reorder, and tag reference images and audio so your characters stay identical across every generated shot — no drift, no re-uploading references per scene.

H3 Segment Prompter — turns a text idea into structured, time-sliced segments using local LLMs (Ollama/Qwen VL) or APIs

H3 Segment Slicer — a lightweight, code-only alternative that slices existing prompts without extra VRAM overhead

H3 Beat Map (powered by Librosa) — detects audio energy shifts and snaps cuts directly to the beat

Set performance roles (on-camera lip-sync vs. voiceover/background) and auto-slice audio to segment length with H3 Track Slice

H3 Timeline — organize, duplicate, lock, and refine individual clips

H3 Shot Dispatcher — manage queue generation via manual index, next-pending, or range modes

Generate fast low-res drafts (0.5MP) to iterate quickly, then run a final high-res pass using the MiniMax H3 Latent Upscaler 3D with FP16 temporal chunking.

H3 Vault — automates queue workflows end to end

H3 Timeline Stitch — stitches all segmented clips with synced audio using background FFmpeg

ComfyUI installed

H3 Planner node suite: https://github.com/AIJigyasa/ComfyUI-H3-Planner

MiniMax H3 base models (free — links in the YouTube tutorial)

Optional: Ollama + Qwen VL for auto-prompting

Creators making AI music videos, lyric videos, or long-form narrative content

Anyone hitting VRAM/length walls trying to generate multi-shot videos locally

Users who want beat-accurate cuts without manual editing