FlashVideo is a high-resolution video generation model based on diffusion architecture. It uses a two-stage approach: first generating low-resolution video from text prompts, then upscaling to high resolution with minimal computation. The system targets efficient inference through flow-based matching and progressive resolution refinement.