MiniMax H3 Original MP3 Lip-Sync Workflow Fixes the issue where input audio is regenerated into pseudo-Chinese vocals. This workflow passes through your original MP3/vocal directly to the final video output while maintaining H3's precise lip-sync.<ul><li>Generation Speed: ~228s for 10-second video (INT8/NVFP4 environment)</li><li>No missing custom nodes needed, standard setup.</li><li><div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/RIN_mp17EBQ" start="0"></iframe></div></li></ul><div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/Bw26PvpolYM?si=FTjbRzlQMz3wGI97" start="0"></iframe></div>

MiniMax H3 - Original Audio Lip-Sync Workflow

# MiniMax H3 Turbo Workflow (Reference to Video + Audio Lip-Sync + Upscalers)This is an optimized ComfyUI workflow for MiniMax H3 Reference-to-Video enhanced with Turbo LoRA and integrated Upscaler pipelines (4x-AnimeSharp &amp; RTX Video Super Resolution).Generated 15s high-quality character dancing/singing videos in ~180s - 200s (RTX 4090 / 5090) with consistent character appearance!---## 📺 YouTube Tutorial &amp; Comparison ReviewWatch the full setup tutorial, comparison with LTX 2.3, and upscaler performance benchmarks:👉<a target="_blank" rel="ugc" href="https://youtu.be/Bw26PvpolYM?si=a603fI8puJtRaIrl">https://youtu.be/Bw26PvpolYM?si=a603fI8puJtRaIrl</a>---## ✨ Key Features &amp; Improvements1. Turbo LoRA Acceleration: Reduces sampling steps from 20+ down to 6-8 steps <code>res_multistep</code>), cutting generation time by more than 50%.2. Audio Lip-Sync &amp; Music Video Support: Fixes previous audio replacement issues. Simply drag-and-drop your custom MP3/WAV into the workflow.3. Built-in Upscaling Pipelines:- RTX Video Super Resolution (VSR): Ultra-fast real-time AI upscaling, ideal for rapid iterations and previews.- 4x-AnimeSharp + ImageScale: Supreme line-art reconstruction and edge clarity for final masterpieces.4. Strong Character Consistency: Reference image identity is preserved throughout the 15-second performance.---## 🛠️ Required Models &amp; Custom Nodes### 1. Base Models (MiniMax H3)Place in their respective standard ComfyUI folders <code>models/unet</code>, <code>models/clip</code>, <code>models/vae</code>):- UNet: <code>minimax_h3_ref2va_pruned_int8_convrot.safetensors</code> (or fp8/fp16)- Text Encoder: <code>qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors</code>- Video VAE: <code>minimax_h3_video_vae_fp16.safetensors</code>- Audio VAE: <code>minimax_h3_audio_vae_fp32.safetensors</code>### 2. Turbo LoRA- <code>minimax_h3_turbo_v4_step600_ema_pruned_comfyui.safetensors</code>- Download from: [Hugging Face (drbaph)](<a target="_blank" rel="ugc" href="https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/tree/main">https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/tree/main</a>)- Place into: <code>ComfyUI/models/loras/</code>### 3. Custom NodesRun the following in <code>ComfyUI/custom_nodes/</code>:```bashgit clone [<a target="_blank" rel="ugc" href="https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo](https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo)">https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo](https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo)</a>git clone [<a target="_blank" rel="ugc" href="https://github.com/Larryvrh/ComfyUI-H3-MultiShot](https://github.com/Larryvrh/ComfyUI-H3-MultiShot)">https://github.com/Larryvrh/ComfyUI-H3-MultiShot](https://github.com/Larryvrh/ComfyUI-H3-MultiShot)</a>

Files with SHA256: 190e9b677b5da632c16b321cd4ce336b816e07505127988a989d2779c6f666e1