Key Features

• Synchronized audio-video generation from text, image, or audio inputs in a single model pass
• Native 4K resolution output at up to 50 FPS for clips up to 20 seconds
• 19B parameter architecture (14B video + 5B audio) built on DiT foundation
• LoRA fine-tuning support with official trainer package included
• FP8 quantization for efficient inference on consumer GPUs
• Multiple pipeline modes: text-to-video, image-to-video, audio-to-video, retake, and distilled
• ComfyUI integration for visual workflow-based generation