Few-Shot-Based Modular Image-to-Video Adapter for Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Zhenhao, Yi, Shaohan, Liu, Zheng, Gao, Leonartinus, Le, Minh Ngoc, Ling, Ambrose, Wang, Zhuoran, Islam, Md Amirul, Chi, Zhixiang, Yu, Yuanhao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912796685369344
author Li, Zhenhao
Yi, Shaohan
Liu, Zheng
Gao, Leonartinus
Le, Minh Ngoc
Ling, Ambrose
Wang, Zhuoran
Islam, Md Amirul
Chi, Zhixiang
Yu, Yuanhao
author_facet Li, Zhenhao
Yi, Shaohan
Liu, Zheng
Gao, Leonartinus
Le, Minh Ngoc
Ling, Ambrose
Wang, Zhuoran
Islam, Md Amirul
Chi, Zhixiang
Yu, Yuanhao
contents Diffusion models (DMs) have recently achieved impressive photorealism in image and video generation. However, their application to image animation remains limited, even when trained on large-scale datasets. Two primary challenges contribute to this: the high dimensionality of video signals leads to a scarcity of training data, causing DMs to favor memorization over prompt compliance when generating motion; moreover, DMs struggle to generalize to novel motion patterns not present in the training set, and fine-tuning them to learn such patterns, especially using limited training data, is still under-explored. To address these limitations, we propose Modular Image-to-Video Adapter (MIVA), a lightweight sub-network attachable to a pre-trained DM, each designed to capture a single motion pattern and scalable via parallelization. MIVAs can be efficiently trained on approximately ten samples using a single consumer-grade GPU. At inference time, users can specify motion by selecting one or multiple MIVAs, eliminating the need for prompt engineering. Extensive experiments demonstrate that MIVA enables more precise motion control while maintaining, or even surpassing, the generation quality of models trained on significantly larger datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2512_20000
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Few-Shot-Based Modular Image-to-Video Adapter for Diffusion Models
Li, Zhenhao
Yi, Shaohan
Liu, Zheng
Gao, Leonartinus
Le, Minh Ngoc
Ling, Ambrose
Wang, Zhuoran
Islam, Md Amirul
Chi, Zhixiang
Yu, Yuanhao
Computer Vision and Pattern Recognition
Diffusion models (DMs) have recently achieved impressive photorealism in image and video generation. However, their application to image animation remains limited, even when trained on large-scale datasets. Two primary challenges contribute to this: the high dimensionality of video signals leads to a scarcity of training data, causing DMs to favor memorization over prompt compliance when generating motion; moreover, DMs struggle to generalize to novel motion patterns not present in the training set, and fine-tuning them to learn such patterns, especially using limited training data, is still under-explored. To address these limitations, we propose Modular Image-to-Video Adapter (MIVA), a lightweight sub-network attachable to a pre-trained DM, each designed to capture a single motion pattern and scalable via parallelization. MIVAs can be efficiently trained on approximately ten samples using a single consumer-grade GPU. At inference time, users can specify motion by selecting one or multiple MIVAs, eliminating the need for prompt engineering. Extensive experiments demonstrate that MIVA enables more precise motion control while maintaining, or even surpassing, the generation quality of models trained on significantly larger datasets.
title Few-Shot-Based Modular Image-to-Video Adapter for Diffusion Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.20000