Reproduce “Know, then Draw: A New Learning Paradigm for Visual Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yucheng Mao, Chongjian Ge, Zhuowen Tu, Hao Tan, Hanwen Jiang
The authors introduce a two-stage text-to-image system: an autoregressive (AR) model predicts visual tokens from a prompt, then diffusion renders pixels from them. They freeze the pretrained AR backbone while training the renderer. A roughly 3-billion-parameter model trained with 6,600 H100 GPU-hours scores 0.86 on GenEval and 87.78 on DPG-Bench, supporting this recipe but not universal superiority or optimal compute allocation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.