Although text-to-image (T2I) models have recently thrived as visual generative priors, their reliance on high-quality text-image pairs makes scaling up expensive. We argue that grasping the cross-modality alignment is not a necessity for a sound visual generative prior, whose focus should be on texture modeling. Such a philosophy inspires us to study image-to-image (I2I) generation, where models can learn from in-the-wild images in a self-supervised manner. We first develop a pure vision-based training framework, Lumos, and confirm the feasibility and the scalability of learning I2I models. We then find that, as an upstream task of T2I, our I2I model serves as a more foundational visual prior and achieves on-par or better performance than existing T2I models using only 1/10 text-image pairs for fine-tuning. We further demonstrate the superiority of I2I priors over T2I priors on some text-irrelevant visual generative tasks, like image-to-3D and image-to-video.
@inproceedings{Lumos2024,
title = {Learning Visual Generative Priors without Text},
author = {Ma, Shuailei and Zheng, Kecheng and Wei, Ying and Wu, Wei and Lu, Fan and Zhang, Yifei and Xie, Chen-Wei and Gong, Biao and Zhu, Jiapeng and Shen, Yujun},
journal = {arXiv},
year = {2024}
}