ContentV: Efficient Training of Video Generation Models with Limited Compute

2025-06-05

AI Summary

To address the prohibitively high training costs of video generation models, this paper introduces ContentV—a text-to-video diffusion model with 8 billion parameters. Methodologically, ContentV features: (1) a minimal temporal extension architecture that reuses a pre-trained image diffusion model, eliminating the need for de novo training; (2) a multi-stage curriculum training strategy based on flow matching to accelerate convergence; and (3) a human-annotation-free RLHF framework for quality optimization, leveraging synthetic feedback to enhance video fidelity and temporal consistency. Trained exclusively on 256×64GB NPUs for four weeks, ContentV achieves a state-of-the-art VBench score of 85.14—marking a substantial reduction in computational overhead. The model weights and training code are fully open-sourced.

Technology Category

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

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent advances in video generation demand increasingly efficient training recipes to mitigate escalating computational costs. In this report, we present ContentV, an 8B-parameter text-to-video model that achieves state-of-the-art performance (85.14 on VBench) after training on 256 x 64GB Neural Processing Units (NPUs) for merely four weeks. ContentV generates diverse, high-quality videos across multiple resolutions and durations from text prompts, enabled by three key innovations: (1) A minimalist architecture that maximizes reuse of pre-trained image generation models for video generation; (2) A systematic multi-stage training strategy leveraging flow matching for enhanced efficiency; and (3) A cost-effective reinforcement learning with human feedback framework that improves generation quality without requiring additional human annotations. All the code and models are available at: https://contentv.github.io.

Problem

• Efficient training of video generation models with limited compute
• Achieving high-quality video generation from text prompts
• Reducing computational costs without sacrificing performance

Innovation

• Reuse pre-trained image models for video
• Multi-stage training with flow matching
• Cost-effective RLHF without extra annotations