StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text

2024-03-21

arXiv.org

AI Summary

Existing text-to-video diffusion models struggle to generate long videos, suffering from hard cuts, scene drift, and motion freezing. This paper introduces the first framework for fluent generation of ultra-long videos (80–1200+ frames). Methodologically, it adopts a diffusion-driven autoregressive architecture integrated with a stochastic resampling fusion strategy. Key contributions include: (1) a Conditional Attention Module (CAM) that models cross-frame temporal dependencies to ensure short-term consistency; (2) an Appearance Preservation Module that anchors high-level features to maintain long-term scene stability; and (3) a stochastic mixing mechanism enabling seamless collaboration between enhancers and the autoregressive diffusion process. Evaluated on multi-scale benchmarks, our approach significantly outperforms prior methods—achieving, for the first time, industrial-grade continuous video synthesis with arbitrary length, high motion dynamics, and zero hard cuts.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Text-to-video diffusion models enable the generation of high-quality videos that follow text instructions, making it easy to create diverse and individual content. However, existing approaches mostly focus on high-quality short video generation (typically 16 or 24 frames), ending up with hard-cuts when naively extended to the case of long video synthesis. To overcome these limitations, we introduce StreamingT2V, an autoregressive approach for long video generation of 80, 240, 600, 1200 or more frames with smooth transitions. The key components are:(i) a short-term memory block called conditional attention module (CAM), which conditions the current generation on the features extracted from the previous chunk via an attentional mechanism, leading to consistent chunk transitions, (ii) a long-term memory block called appearance preservation module, which extracts high-level scene and object features from the first video chunk to prevent the model from forgetting the initial scene, and (iii) a randomized blending approach that enables to apply a video enhancer autoregressively for infinitely long videos without inconsistencies between chunks. Experiments show that StreamingT2V generates high motion amount. In contrast, all competing image-to-video methods are prone to video stagnation when applied naively in an autoregressive manner. Thus, we propose with StreamingT2V a high-quality seamless text-to-long video generator that outperforms competitors with consistency and motion. Our code will be available at: https://github.com/Picsart-AI-Research/StreamingT2V

Problem

• Generates long videos with smooth transitions from text
• Prevents forgetting initial scene in long video synthesis
• Enables consistent enhancement for infinitely long videos

Innovation

• Autoregressive long video generation with smooth transitions
• Conditional attention module for consistent chunk transitions
• Appearance preservation module prevents forgetting initial scene