This workflow turns a single still image into a short, coherent video clip with synchronized stereo audio using MiniMax H3. At its core is the MiniMax H3 generation node (id: 4c314f31-ecda-4b08-ae98-faaba1bf613f), which performs a single forward pass to synthesize both video frames and audio that match the scene. The image you provide through LoadImage sets the subject and overall style; you can optionally add short text guidance inside the MiniMax H3 node to suggest motion, pacing, or mood.

To help the model run efficiently and preserve composition, the workflow routes the input through GetImageSize and a Use Image Size group that keeps aspect ratio intact. ResolutionSelector and ImageScaleToTotalPixels work together to pick a target size that fits MiniMax H3’s recommended megapixel budgets and the multiple-of-32 constraint required by the model. The generated clip is then written out by SaveVideo as a single MP4 containing the video stream (and the model’s generated audio, if enabled), ready for quick review or sharing. Model assets used by the generator include the video VAE (minimax_h3_video_vae_fp16.safetensors) and the audio VAE (minimax_h3_audio_vae_fp32.safetensors).