Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models

Lvmin Zhang, Shengqu Cai, Muyang Li, Gordon Wetzstein, Maneesh Agrawala

Keywords: video diffusion models, autoregressive video models

TL;DR: Methods for packing frame context and preventing drifting phenomenon in next-frame-prediction video models

Abstract: We present a neural network structure, FramePack, to train next-frame (or next-frame-section) prediction models for video generation. FramePack compresses input frame contexts with frame-wise importance so that more frames can be encoded within a fixed context length, with more important frames having longer contexts. The frame importance can be measured using time proximity, feature similarity, or hybrid metrics. The packing method allows for inference with thousands of frames and training with relatively large batch sizes. We also present drift prevention methods to address observation bias (error accumulation), including early-established endpoints, adjusted sampling orders, and discrete history representation. Ablation studies validate the effectiveness of the anti-drifting methods in both single-directional video streaming and bi-directional video generation. Finally, we show that existing video diffusion models can be finetuned with FramePack, and analyze the differences between different packing schedules.

Primary Area: Deep learning (e.g., architectures, generative models, optimization for deep networks, foundation models, LLMs)

Submission Number: 11170