Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models | Read Paper on Bytez