This ComfyUI workflow turns a single text prompt into a complete short video with native stereo audio using MiniMax H3, an omni‑modal generative model. The core generation is handled by the H3 node (4c314f31-ecda-4b08-ae98-faaba1bf613f), which jointly synthesizes visuals, voice, sound effects, and music in one forward pass. ResolutionSelector chooses a valid width/height pair (multiples of 32) for stable decoding, while SaveVideo writes the final container with embedded audio. A MarkdownNote on the canvas provides quick tips and model links. Typical outputs are ~15 seconds at 24 fps and can reach up to 2K resolution—ideal for rapid concept visualization and audio‑synced storytelling without any file inputs.

Under the hood, the workflow uses MiniMax’s paired VAEs to decode both modalities: a video VAE (minimax_h3_video_vae_fp16.safetensors) for frames and an audio VAE (minimax_h3_audio_vae_fp32.safetensors) for stereo sound. By routing the H3 node’s unified output directly to SaveVideo, you get a single MP4 (or chosen container) that’s already synchronized, with no manual audio mixing required. Because the model generates image and audio together, prompt wording guides both: you can describe scenes, pacing, and audio cues (e.g., “soft piano music,” “crackling fireplace,” or “narrator: …”) to shape timing and tone in one pass.