The Year in AI Papers
The year/Independent research

Paper 2511.15700

First Frame Is the Place to Go for Video Content Customization

Jingxi Chen, Zongxia Li, Zhichao Liu, Guangyao Shi, Xiyang Wu, Fuxiao Liu, Cornelia Fermuller, Brandon Y. Feng, Yiannis Aloimonos

Published
Nov 2025
Research lab
Independent
Citations
8
GitHub
194 stars

01 In brief

Summary

This paper introduces FFGo, a lightweight add-on for video generation models that enables multi-reference video content customization without architectural changes or large-scale fine-tuning.

The authors discover that pre-trained video models treat the first frame as a conceptual memory buffer, storing visual entities for later reuse.

FFGo leverages this by using a mixed-subject first frame and a unique transition phrase to invoke the model's innate ability to fuse subjects and perform scene transitions.

The pipeline includes dataset curation using VLMs (Gemini-2.5-Pro) and SAM 2 to create training data, followed by few-shot LoRA adaptation with only 20-50 examples.

Experiments on Wan2.2-I2V-A14B show FFGo outperforms state-of-the-art baselines (VACE, SkyReels-A2) in user studies across diverse applications like robotic manipulation, driving simulation, and filmmaking, while preserving the base model's pre-trained knowledge.

The method supports up to five reference subjects and achieves 81.2% top-1 ranking in user evaluations.

02 From the paper

Abstract

What role does the first frame play in video generation models? Traditionally, it's viewed as the spatial-temporal starting point of a video, merely a seed for subsequent animation. In this work, we reveal a fundamentally different perspective: video models implicitly treat the first frame as a conceptual memory buffer that stores visual entities for later reuse during generation. Leveraging this insight, we show that it's possible to achieve robust and generalized video content customization in diverse scenarios, using only 20-50 training examples without architectural changes or large-scale finetuning. This unveils a powerful, overlooked capability of video generation models for reference-based video customization.