生成模型与扩散 · CVPR2026 ·2026.06.03

ThinkGen:面向视觉生成的泛化思维

ThinkGen: Generalized Thinking for Visual Generation

Siyu Jiao、Yiheng Lin、Yujie Zhong、Qi She、Wei Zhou、Xiaohan Lan 等 12 人

与你有关的部分

把 MLLM 的思维链推理作为扩散生成的指令来源,并用可分离的 GRPO 交替训练 MLLM 与 DiT,为视觉生成提供了一条可跨场景泛化的「先推理后生成」训练范式,可直接迁移到文生图、指令式编辑等条件生成任务。

原文依据 ThinkGen employs a decoupled architecture comprising a pretrained MLLM and a Diffusion Transformer (DiT)
可迁移到 text-to-image generationinstruction-based image generationcontrollable generation

摘要译文

由模型翻译 · 术语统一 · 请对照原文核对关键结论

多模态大语言模型(MLLM)的最新进展表明,思维链(CoT)推理能够为复杂理解任务提供系统化的解决方案。然而,将其扩展到生成任务的研究仍处于起步阶段,并且受限于场景特定的机制,难以实现泛化与适配。在本工作中,我们提出 ThinkGen,这是首个在各种生成场景中显式利用 MLLM 思维链推理的思维驱动视觉生成框架。ThinkGen 采用由预训练 MLLM 与扩散 Transformer(DiT)组成的解耦架构,其中 MLLM 根据用户意图生成定制化指令,DiT 在这些指令的引导下生成高质量图像。我们进一步提出一种可分离的基于 GRPO 的训练范式(SepGRPO),在 MLLM 与 DiT 模块之间交替进行强化学习。这一灵活的设计使得跨多样数据集的联合训练成为可能,从而为广泛的生成场景提供有效的思维链推理。大量实验表明,ThinkGen 在多个生成基准上取得了稳健的最优性能。

英文原文摘要 1148 字符

Recent progress in Multimodal Large Language Models (MLLMs) demonstrates that Chain-of-Thought (CoT) reasoning enables systematic solutions to complex understanding tasks. However, its extension to generation tasks remains nascent and limited by scenario-specific mechanisms that hinder generalization and adaptation. In this work, we present ThinkGen, the first think-driven visual generation framework that explicitly leverages MLLM's CoT reasoning in various generation scenarios. ThinkGen employs a decoupled architecture comprising a pretrained MLLM and a Diffusion Transformer (DiT), wherein the MLLM generates tailored instructions based on user intent, and DiT produces high-quality images guided by these instructions. We further propose a separable GRPO-based training paradigm (SepGRPO), alternating reinforcement learning between the MLLM and DiT modules. This flexible design enables joint training across diverse datasets, facilitating effective CoT reasoning for a wide range of generative scenarios. Extensive experiments demonstrate that ThinkGen achieves robust, state-of-the-art performance across multiple generation benchmarks.

标签

会议 CVPR2026
页码 14713-14723
Bibkey
来源 CVF Open Access