BachVid: Training-Free Video Generation with Consistent Background and Character

2025-10-24

AI Summary

Generating multiple videos with consistent characters and backgrounds remains challenging: existing approaches rely either on reference images or extensive training, and often ensure only character consistency. This paper introduces BachVid—the first training-free, reference-free framework for consistent video generation. Its core innovation lies in analyzing the attention mechanisms and intermediate features of Diffusion Transformers (DiTs) to automatically extract foreground masks, identify cross-frame correspondence points, and cache/reuse identity-relevant intermediate variables to inject consistency into new generations. BachVid jointly models inter-video consistency for both foreground subjects and background scenes. Experiments demonstrate that, under zero-training conditions, BachVid significantly improves visual consistency across multiple generated videos—achieving stable, efficient performance while eliminating reliance on external supervision signals or manually designed modular components.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Diffusion Transformers (DiTs) have recently driven significant progress in text-to-video (T2V) generation. However, generating multiple videos with consistent characters and backgrounds remains a significant challenge. Existing methods typically rely on reference images or extensive training, and often only address character consistency, leaving background consistency to image-to-video models. We introduce BachVid, the first training-free method that achieves consistent video generation without needing any reference images. Our approach is based on a systematic analysis of DiT's attention mechanism and intermediate features, revealing its ability to extract foreground masks and identify matching points during the denoising process. Our method leverages this finding by first generating an identity video and caching the intermediate variables, and then inject these cached variables into corresponding positions in newly generated videos, ensuring both foreground and background consistency across multiple videos. Experimental results demonstrate that BachVid achieves robust consistency in generated videos without requiring additional training, offering a novel and efficient solution for consistent video generation without relying on reference images or additional training.

Problem

• Achieving consistent character and background across multiple generated videos
• Eliminating dependency on reference images or extensive training
• Leveraging attention mechanisms for foreground-background consistency

Innovation

• Training-free video generation without reference images
• Leveraging DiT attention for foreground masks and matching
• Injecting cached variables to ensure foreground-background consistency