# 🚀 MiniMax H3 Motion Director: VRAM-Optimized Multi-Shot Suite (12GB Friendly)UPDATE: There is a second version available to download in the drop-down menu that is compatible with COmfyUI version 0.35.This is not a replacement for the 30-second MiniMax H3 multi-shot workflow, since that workflow still allows you to make 30-second generations in just 14 minutes.This here allows you to do more seamless tracking shots at up to a minute (or 60 seconds) in just 35 minutes of generation time, all thanks to j955229's amazing H3 Motion Director node.For those who want to do more complex video production, this workflow enables you to do so without the time-constraints of the 30-second workflow. This is also useful if you want to do tests of longer video generations without burning tons of credits on premium AI platforms. Give your gens a test-run locally before opting to use higher resolutions or more expensive models.This is basically an advanced, production-ready ComfyUI workflow designed specifically for long-form, multi-segment video generation on consumer-grade hardware. Tested and benchmarked to cleanly render a 77-second continuous sequence (1,800+ frames) without Out-Of-Memory (OOM) crashes, while achieving a 30% speedup over raw generation architectures. --- ##💡 The Core Innovation: "Unified Boolean Logic Switch" Most 12GB workflows force you to choose between slow, high-fidelity 20-step renders or permanently soft 8-step distilled outputs.This canvas utilizes a custom Unified Boolean Toggle that syncs model weights and scheduler steps across your pipeline simultaneously:* Normal Mode (Unchecked): Runs full-precision base diffusion models at 20 steps for maximum micro-detail preservation.* Turbo Mode (Checked): Instantly drops the scheduler to 8 steps and scales down the LoRA distillation weights in perfect synchronization to fly through complex temporal tracking layouts. By routing this logic through the native <code>strength_model</code> parameter, this workflow completely avoids the notorious <code>COMFY_MATCHTYPE_V3</code> wildcard validation error common in advanced scripting pipelines.## 📦 Required Custom Nodes / DependenciesEnsure your ComfyUI is updated to v0.34 or later before importing. Download the following extensions via ComfyUI Manager:* ComfyUI-MiniMax-H3-Motion-Director (Main Generation Engine)* comfyui-kjnodes (For SageAttention optimization layer and INT constants)* diffaid-patches (For the MiniMax H3 Diff-Aid Sparse Patch stability layer)* comfyui-h3-multishot (For advanced text/vision clip handling)---## 🚀 Setup &amp; Model RequirementsPlace your models in the corresponding directories inside your ComfyUI root:1. UNET/Diffusion: <code>minimax_h3_fl2va_pruned_int8_convrot.safetensors</code> -&gt; <code>models/diffusion_models/</code>2. Text Encoder/CLIP: <code>qwen3vl_32b_minimax_h3_int4_convrot.safetensors</code> -&gt; <code>models/clip/</code>3. Turbo Weights: <code>minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors</code> -&gt; <code>models/loras/</code>4. Video/Audio VAE: Both respective H3 Safetensors files -&gt; <code>models/vae/</code>---## 🎨 How to Use This Template1. Drag and drop the provided <code>.json</code> file onto your blank ComfyUI canvas.2. Update the timeline segments inside the MiniMaxH3MotionDirector node with your narrative scripts (The template file includes a pre-configured Arctic Hunter video using the text-to-video option within the H3 Motion Director).3. Flip the Master Turbo Toggle primitive checkbox to your preferred rendering speed/fidelity tier.4. Hit Queue Prompt and enjoy stable, crash-free long-form cinematic workflows!

MiniMax H3 60-Second Seamless Video Workflow For 12GB GPUs @ 35-Min Gen Times

An advanced, production-ready ComfyUI workflow designed specifically for long-form, multi-segment video generation on consumer-grade hardware. Tested and benchmarked to cleanly render a 77-second continuous sequence (1,800+ frames) without Out-Of-Memory (OOM) crashes, while achieving a 30% speedup over raw generation architectures.

Files with SHA256: ada956316efccf8cec472bfda4ef16a6938b1f2d7add4900abea41a828109ace