Goku: Flow Based Video Generative Foundation Models

TL;DR

Goku introduces a new flow-based model family for high-quality joint image-and-video generation, achieving state-of-the-art results through innovative architecture, data pipeline, and training infrastructure.

Contribution

The paper presents Goku, a novel flow-based model family that advances joint image-and-video generation with superior performance and scalable training infrastructure.

Findings

Achieves 0.76 on GenEval for text-to-image generation

Scores 83.65 on DPG-Bench for text-to-image tasks

Reaches 84.85 on VBench for text-to-video tasks

Abstract

This paper introduces Goku, a state-of-the-art family of joint image-and-video generation models leveraging rectified flow Transformers to achieve industry-leading performance. We detail the foundational elements enabling high-quality visual generation, including the data curation pipeline, model architecture design, flow formulation, and advanced infrastructure for efficient and robust large-scale training. The Goku models demonstrate superior performance in both qualitative and quantitative evaluations, setting new benchmarks across major tasks. Specifically, Goku achieves 0.76 on GenEval and 83.65 on DPG-Bench for text-to-image generation, and 84.85 on VBench for text-to-video tasks. We believe that this work provides valuable insights and practical advancements for the research community in developing joint image-and-video generation models.