This workflow is a ultra-fast MiniMax H3 Reference-to-Video pipeline accelerated by Turbo LoRA (6~8 Steps). It enables you to generate a 15-second high-quality reference video with lip-sync in under 200 seconds (tested on RTX 5090). (MiniMax H3のReference-to-VideoをTurbo LoRAで高速化し、15秒動画を約200秒弱で生成できる爆速ワークフローです。)

【Required Models / 必要なモデル】

• Reference Diffusion Model:

  • minimax_h3_ref2va_pruned_int8_convrot.safetensors (or fp8 version)

  • Save to: ComfyUI/models/diffusion_models/ (or models/unet/)

• Turbo LoRA Model:

  • minimax_h3_turbo_v4_step600_ema_prun...safetensors

  • Save to: ComfyUI/models/loras/

• Text Encoders & VAE:

  • Standard MiniMax H3 CLIP/Text Encoders in models/clip/ and VAEs in models/vae/.

【Custom Nodes Installation / カスタムノード導入手順】

Open your command prompt in the ComfyUI/custom_nodes directory and run the following commands: (ComfyUIの custom_nodes フォルダを開き、アドレスバーに cmd と入力してEnterを押した後、以下のコマンドを実行してください)

command1：git clone https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo

command2：git clone https://github.com/jlucasmcrell/ComfyUI-H3-Multishot

Restart ComfyUI after cloning. (クローン完了後、ComfyUIを再起動してください)

【Recommended Settings / 推奨設定】

• Sampler: MiniMax-H3 Turbo Sampler

• Scheduler: simple

• Steps: 6 - 8

• LoRA Strength: 1.0

• Duration: 15.0 seconds (at 24 FPS)

【Credits】

• MiniMax H3 Turbo Node: Larryvrh

• MultiShot Node: jlucasmcrell