让AI画图前先“想”一下：连续空间推理
现在的AI画图，你给它一段描述，它直接“画”，中间没有思考过程。这篇论文发现，如果让AI像写作文一样先“想”几步再画，效果反而变差——因为语言是离散的，而图像是连续的，两者不兼容。于是研究者让AI在“连续空间”里思考，相当于在画布上直接打草稿，而不是先写文字再翻译成图。结果，图像质量、指令遵循和构图准确性都提升了，在GenEval基准上比之前最好的模型高出7个百分点，而且思考时间越长，效果越好。这不是你明天就能用的工具，但它指向一个方向：AI画图的下一个突破，可能不是画得更快，而是想得更深。
📄 原文摘要(英文)
Unified Multimodal Models (UMMs) achieve remarkable success on diverse multimodal understanding and generation tasks, yet they still struggle to effectively leverage language-based chain-of-thought reasoning for image generation. As a result, existing reasoning-augmented UMMs often fail to inherit the test-time scaling and fine-grained controllability observed in large language models. We attribute this failure to a fundamental modality mismatch: reasoning is performed in a discrete token space, whereas image generation operates in a continuous space. To bridge this gap, we introduce a Latent Reasoning framework in Continuous space (LARC), which alleviates the token-level bottleneck by reasoning over continuous hidden states. Specifically, the training framework consists of two stages: (i) curriculum supervised fine-tuning (SFT), and (ii) information-gain reinforcement learning (RL). First, the curriculum SFT stage gradually converts token-level reasoning into latent reasoning. Subsequently, the information-gain RL stage self-evolves latent reasoning without ground truth supervision. Experiments across diverse text-to-image generation benchmarks show that LARC consistently improves generation quality, prompt following, and compositional fidelity. Notably, LARC achieves state-of-the-art performance on GenEval with a 7%p gain over BAGEL and demonstrates clear test-time scaling behavior, outperforming token-based generation under both parallel and sequential test-time scaling strategies.