Newest papers

Text-to-Image Generative Modeling Without Language Pretraining

Keshigeyan Chandrasegaran, Yue Zhao, Armin W. Thomas, Ehsan Adeli, Jiajun Wu, Juan Carlos Niebles, Li Fei-Fei, Michael Poli · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26863 · cited by 0 works (OpenAlex) · open-access PDF

Korean

현대 텍스트-to-Image (T2I) 시스템은 텍스트 조건화 이미지를 생성하기 위해 인터넷 규모의 언어 프리 트레이닝을 사용합니다. 우리는 강력한 T2I 생성에 필요한 언어 프리 트레이닝이 필요한지 여부를 묻습니다. 우리는 초기 융합 멀티모델 프리 트레이닝 모델의 가족인 Halcyon을 소개합니다. GPIC에서 처음부터 전적으로 프리 트레이닝 된 100M 이미지-텍스트 쌍의 데이터 세트인 Halcyon-0.52B는 LlamaGen-XL/8 (48.0 vs. 63.3 gFD)를 18.2$\times$ 적은 프리 트레이닝 컴퓨팅을 제공합니다. 또한 Halcyon-1.6B는 JiT-T2I-XXL/8++ (35.3 vs. 41.0F gD

Machine translation (Meta M2M100); the English original follows.

Abstract

Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2$\times$ less pretraining compute. Further, Halcyon-1.6B outperforms JiT-T2I-XXL/8++ (35.3 vs. 41.0 gFD) with 275$\times$ less pretraining compute. Controlled experiments show that text-only pretraining does not improve T2I generation. Halcyon also learns image captioning by replacing caption dropout with I2T supervision, maintaining competitive T2I generation quality under the same pretraining budget. We show that strong T2I systems can be trained from scratch without internet-scale language pretraining. Code and models are available at https://cs.stanford.edu/~keshik/halcyon

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Closed captioning, Computer science, Artificial intelligence, Scratch, Natural language processing, Language model

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange