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
Moderne Text-zu-Bild-Systeme (T2I) verwenden Internet-Skala-Sprachvorbereitung für textbedingte Bildgeneration. Wir fragen, ob Sprachvorbereitung für starke T2I-Generation erforderlich ist. Wir stellen Halcyon, eine Familie von frühen Fusion multimodalen autoregressiven Modellen vorbereitet vollständig von Grund auf GPIC, ein Datensatz von 100M Bild-Text-Paare. Halcyon-0.52B übertrifft LlamaGen-XL/8 (48.0 vs. 63.3 gFD) mit 18,2$\times$ weniger vorbereitenden Computern. Weiterhin übertrifft Halcyon-1.6B JiT-T2I-XXL/8++ (35.3 vs. 41.0F gD) mit 275$\times weniger vorbereitenden
Machine translation (Meta M2M100); the English original follows.
Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2$\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
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26863, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange