<h1 id="minimax-h3-minimal-i2v-+-audio-workflow-for-comfyui">MiniMax H3 – Minimal I2V + Audio Workflow for ComfyUI</h1>A compact MiniMax H3 image-to-video workflow for ComfyUI with synchronized audio generation.This package contains a tested workflow JSON, documentation, and direct model download references. Model weights are not included. Please download them from the original Comfy-Org repository.<hr /><h2 id="features">Features</h2><ul><li>Image-to-video from a single starting image</li><li>Joint video + audio generation</li><li>124 frames at 24 fps (about 5.17 seconds)</li><li><code>res_multistep</code> sampler</li><li>20 steps</li><li>MiniMax H3 sigma shift:<ul><li>Video: 12</li><li>Audio: 3</li></ul></li><li>Standard ComfyUI VAE loaders</li><li>Uses the official Comfy-Org Audio VAE with latent statistics included</li><li>VideoHelperSuite output to H.264 MP4 with audio</li></ul><hr /><h2 id="required-model-files">Required model files</h2><h3 id="1.-diffusion-model">1. Diffusion model</h3><code>minimax_h3_fl2va_pruned_int8_convrot.safetensors</code>Direct download: <a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors?download=true">Download diffusion model</a>Place in: <code>ComfyUI/models/diffusion_models/</code> or your configured <code>diffusion_models</code> folder.<hr /><h3 id="2.-text-encoder">2. Text encoder</h3><code>qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors</code>Direct download: <a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors?download=true">Download text encoder</a>Place in: <code>ComfyUI/models/text_encoders/</code> or your configured <code>text_encoders</code> folder.<hr /><h3 id="3.-video-vae">3. Video VAE</h3><code>minimax_h3_video_vae_fp16.safetensors</code>Direct download: <a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors?download=true">Download video VAE</a>Place in: <code>ComfyUI/models/vae/</code> or your configured <code>vae</code> folder.<hr /><h3 id="4.-audio-vae">4. Audio VAE</h3><code>minimax_h3_audio_vae_fp32.safetensors</code>Direct download: <a target="_new" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors?download=true">Download audio VAE</a>Place in: <code>ComfyUI/models/vae/</code> or your configured <code>vae</code> folder.<hr /><h2 id="required-custom-node">Required custom node</h2><ul><li>ComfyUI-VideoHelperSuite</li></ul>MiniMax H3 model / conditioning nodes used by this workflow are ComfyUI core nodes in a ComfyUI build with MiniMax H3 support.<hr /><h2 id="sample-prompt-english-speech">Sample prompt – English speech</h2><pre><code>A woman looks directly at the camera and speaks naturally in English: "Hello! I'm testing the audio generation. Can you hear my voice clearly?" Her voice is clear and easy to understand. Natural synchronized English speech, accurate lip movements, and subtle room ambience.</code></pre><hr /><h2 id="workflow-settings">Workflow settings</h2><ul><li> Resolution in included tested sample: 512 × 512 </li><li> Length: 124 frames </li><li> Frame rate: 24 fps </li><li> Sampler: <code>res_multistep</code> </li><li> Scheduler: <code>normal</code> </li><li> Steps: 20 </li><li> Denoise: 1.0 </li><li> Video sigma shift: 12 </li><li> Audio sigma shift: 3 </li></ul>You can change the prompt, source image, and resolution as needed.<hr /><h2 id="included-files">Included files</h2><ul><li><code>MiniMax_H3_I2V_test.json</code> <ul><li> Tested 512×512 English speech example </li><li> Source image connected to <code>first_frame</code> </li></ul></li><li><code>README.md</code> </li><li><code>CIVITAI_DESCRIPTION.md</code> </li><li><code>MODEL_DOWNLOADS.md</code> </li><li><code>MANIFEST.txt</code> </li></ul><hr /><h2 id="important-note">Important note</h2>Use the official Comfy-Org Audio VAE listed above.A differently packaged Audio VAE that lacks <code>latents_mean</code> / <code>latents_std</code>, or uses unfused <code>weight_g</code> / <code>weight_v</code> decoder weights, may decode to:<ul><li> silence </li><li> noise </li><li> a constant tone </li></ul>This package does not redistribute model weights. Please follow the licenses and terms of the original model repositories.

Files with SHA256: c8320328fd27b0947308c752dbe5325ba1e85994435a7b2fa80f36f75dc2c33e