Animated Stickers: Bringing Stickers to Life with Video Diffusion

TL;DR

This paper presents a novel video diffusion model for animated stickers that generates vivid, motion-rich animations from text prompts and static images, using a specialized two-stage finetuning process.

Contribution

It introduces a two-stage finetuning pipeline with human-in-the-loop strategies to adapt diffusion models for sticker animation, focusing on motion quality while preserving style.

Findings

High-quality animation generated in under one second

Effective adaptation of diffusion models for sticker domain

Improved motion realism through ensemble-of-teachers finetuning

Abstract

We introduce animated stickers, a video diffusion model which generates an animation conditioned on a text prompt and static sticker image. Our model is built on top of the state-of-the-art Emu text-to-image model, with the addition of temporal layers to model motion. Due to the domain gap, i.e. differences in visual and motion style, a model which performed well on generating natural videos can no longer generate vivid videos when applied to stickers. To bridge this gap, we employ a two-stage finetuning pipeline: first with weakly in-domain data, followed by human-in-the-loop (HITL) strategy which we term ensemble-of-teachers. It distills the best qualities of multiple teachers into a smaller student model. We show that this strategy allows us to specifically target improvements to motion quality while maintaining the style from the static image. With inference optimizations, our model is able to generate an eight-frame video with high-quality, interesting, and relevant motion in under one second.