<p>Turn a picture you already made into a moving clip that comes with its own sound. It runs on your own machine in ComfyUI. No API, no credits, nothing to sign up for.</p><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/iBP6nP4BZ84" start="0"></iframe></div><p>17 nodes, all plain ComfyUI. There are no custom node packs to install.</p><h3>What you need</h3><p>Four files, about 42 GB in total, all free from the official Comfy-Org MiniMax-H3 repository on Hugging Face. Every filename below is a direct download link.</p><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors">minimax_h3_fl2va_pruned_int8_convrot.safetensors</a> 21 GB, into <code>ComfyUI/models/diffusion_models/</code></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors">qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors</a> 15.7 GB, into <code>ComfyUI/models/text_encoders/</code></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors">minimax_h3_video_vae_fp16.safetensors</a> 5.2 GB, into <code>ComfyUI/models/vae/</code></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors">minimax_h3_audio_vae_fp32.safetensors</a> 0.6 GB, into <code>ComfyUI/models/vae/</code></li></ul><p>The audio VAE is the one people skip, and it is the one that gives you the sound. Put each file in the folder shown. A file in the wrong folder looks exactly like a missing model.</p><p>You also need ComfyUI 0.30 or newer, and a card with 24 GB or more.</p><h3>The bit most people get wrong</h3><p>Describe what <strong>moves</strong>, not what is in the shot. "she turns her head and smiles, the camera drifts slowly to the left" gives you a clip. "a woman in a field at sunset" gives you a picture that mostly sits there. The model can already see your image. You are telling it what happens next.</p><h3>A free camera move</h3><p>There is a second slot for a last frame. Crop your own picture tighter, hand that crop back as the ending, and you get a smooth push in. No camera nodes, no keyframes.</p><h3>Match the size to your picture</h3><p>The clip has to be the same shape as your image. It does not crop to fit, it squashes, and it will not warn you. Both numbers also have to divide by 32, or you get a tensor shape error that never mentions picture size.</p><p>There is a companion workflow that builds a clip from a prompt alone, no picture needed. It is on my profile. The written walkthrough for both is <a target="_blank" rel="ugc" href="https://forgebun.com/go/civit_h3_workflows">on my portal</a>. Free account, no card.</p>

Files with SHA256: 07eca826b3e29344db55a828d062cb9b02b4f5046ecec7790ca36d20baf3ce0f