< Explain other AI papers

LongAnimation: Long Animation Generation with Dynamic Global-Local Memory

Nan Chen, Mengqi Huang, Yihao Meng, Zhendong Mao

2025-07-03

LongAnimation: Long Animation Generation with Dynamic Global-Local
  Memory

Summary

This paper talks about LongAnimation, a system that helps colorize long animation videos while keeping the colors consistent across many frames. It uses a smart memory system that remembers global and local color information to make sure the colors don’t change unexpectedly throughout the animation.

What's the problem?

The problem is that coloring long animations by hand takes a lot of time and effort. Existing AI methods can color animations but usually only work well for short clips, and they struggle to keep colors consistent over long sequences, causing colors to flicker or change abruptly.

What's the solution?

The researchers created LongAnimation, which combines three main parts: a system to understand reference images and sketches, a dynamic memory that keeps track of color information from the whole animation and the recent frames, and a special reward system that encourages the model to keep colors consistent. This approach helps generate long animations with smooth and stable coloring.

Why it matters?

This matters because it can greatly reduce the work needed in animation production, making it faster and easier to create high-quality colored animations with consistent colors, which is important for movies, games, and other digital media.

Abstract

LongAnimation framework achieves long-term color consistency in animation colorization by integrating a dynamic global-local memory and color consistency reward.