This Custom Node Suite Solves MiniMax H3's Biggest Limit — H3 Planner

In this video, I'll show you how to create multi-minute AI music videos using MiniMax H3 — even on an 8GB or 12GB VRAM GPU like the RTX 3060. I built a custom node suite called H3 Planner that breaks long-form generation into a manageable pipeline: cast management, automated prompt slicing, beat-synced cuts, a full timeline editor, draft-to-final upscaling, and auto-stitching — all inside ComfyUI. This video walks through the entire system and shows a full demo, start to finish.

Featuring H3 Plannergithub.com Workflowbuymeacoffee.com

• 💎 This is a paid workflow — free with AI Jigyasa membership (Builder/Prayog and Studio)

💎 AI Jigyasa Membershipbuymeacoffee.com

Timestamps

00:00Introduction & VRAM Limitations on 8GB/12GB GPUs 00:43Demo Video Preview (AI Rock Metal Music Video) 02:10The Problem with Generating Long-Form AI Videos 04:08How to Install H3 Planner (ComfyUI Manager & GitHub) 05:53Workflow Overview & Node Pack Introduction 07:23Audio Input & Trimming Section 07:54Custom Prompt Input Section (Fable 5.1 / Claude / ChatGPT) 08:28H3 CastBoard Node (Character & Media References) 10:22H3 Project Node (Project Settings, FPS, Draft vs. Final Passes) 11:42H3 Timeline Node (Overview & Segment Planning) 13:12Channel Memberships & Support Tiers Announcement ($1, $6 Builder, $15 Client Tiers) 14:41H3 Segment Prompter Node (LLM & Vision-Based Prompt Slicing) 17:00Setting Audio Roles (Lip-sync vs. Voiceover vs. Background Only) 19:58H3 Segment Slicer Node (Code-Only / Zero-VRAM Alternative) 21:06H3 Beat Map Node (Librosa Beat Detection & Energy Shifts) 22:04Timeline Management (Locking, Redo, Inline, Saved & Merge Settings) 25:37MiniMax Model Setup & VRAM Optimizations (W4A8 & FP4) 26:57H3 Track Slice Node (Audio Slicing per Shot) 27:33H3 Shot Dispatcher Node (Queue Modes: Manual Index, Next Pending, Range) 28:33H3 Pass Gate Node (Draft vs. Final State Tracking) 29:15H3 Timeline Stitch Node (Background FFmpeg Stitching) 29:48Latent Upscaler 3D (FP16 & Temporal Chunking) 31:38H3 Vault Node & Auto-Advance (Queue Automation) 33:20Live Generation Demo (Next Pending, Manual Index & Range Runs) 35:50Legacy Nodes & Whisper Audio Timestamping Option 37:04Membership Recap, Outro & Future Updates (Story Planner)

Installation

Clone the H3 Planner node suite into your ComfyUI custom_nodes folder:

Restart ComfyUI (or install via ComfyUI Manager → search "H3 Planner")
Install dependencies if prompted (Librosa for beat mapping, FFmpeg for stitching)

Model links

diffusion_models-

minimax_h3_fl2va_pruned_int8_convrot.safetensorshuggingface.co For 8gb cardshuggingface.co

• 📂 Place in: ComfyUI/models/diffusion_models/

models/text_encoders

qwen3vl_32b_minimax_h3_nvfp4_awq.safetensorshuggingface.co

• 📂 Place in: ComfyUI/models/text_encoders/

models/vae

minimax_h3_video_vae_fp16.safetensorshuggingface.co For Low Vram (video vae)huggingface.co Comfy-Org/MiniMax-huggingface.co

H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors?download=true

• 📂 Place in: ComfyUI/models/vae/

upscale_models (for final pass)

minimax_h3_latent_upscaler_3d_fp16.safetensorshuggingface.co

• 📂 Place in: ComfyUI/models/upscale_models/

local LLM (optional — for H3 Segment Prompter)

Ollama + Qwen VLollama.com

• 👍 Support the Channel If you found this helpful:

• 👍 Like the video
• 🔔 Subscribe for more AI workflows
• 💬 Comment what you want to see next

#MiniMaxH3#ComfyUI#AIMusicVideo#H3Planner#GenerativeAI#LongFormAI#12GBVRAM#OpenSourceAI