UltraImage：重新思考图像扩散变换器中的分辨率外推

摘要

近期基于扩散变换器的图像生成模型虽能实现高保真度生成，但在超越训练尺度时会出现内容重复与质量下降的问题。本文提出UltraImage这一原理性框架以同时解决这两个难题。通过对位置嵌入进行频域分析，我们发现内容重复源于主导频率的周期性特征——其周期与训练分辨率保持一致。为此，我们引入递归式主导频率校正技术，在分辨率外推后将主导频率约束在单一周期内。此外，我们发现质量下降源于注意力稀释现象，进而提出熵引导的自适应注意力集中机制：通过分配更高的聚焦因子来锐化局部注意力以增强细节表现，同时降低全局注意力模式的聚焦程度以保持结构一致性。实验表明，UltraImage在Qwen-Image和Flux（约4K分辨率）的三种生成场景中均优于现有方法，有效减少重复现象并提升视觉保真度。更值得注意的是，UltraImage仅凭1328p的训练分辨率即可生成高达6K*6K的图像（无需低分辨率引导），展现出卓越的外推能力。项目页面详见https://thu-ml.github.io/ultraimage.github.io/。

English

Recent image diffusion transformers achieve high-fidelity generation, but struggle to generate images beyond these scales, suffering from content repetition and quality degradation. In this work, we present UltraImage, a principled framework that addresses both issues. Through frequency-wise analysis of positional embeddings, we identify that repetition arises from the periodicity of the dominant frequency, whose period aligns with the training resolution. We introduce a recursive dominant frequency correction to constrain it within a single period after extrapolation. Furthermore, we find that quality degradation stems from diluted attention and thus propose entropy-guided adaptive attention concentration, which assigns higher focus factors to sharpen local attention for fine detail and lower ones to global attention patterns to preserve structural consistency. Experiments show that UltraImage consistently outperforms prior methods on Qwen-Image and Flux (around 4K) across three generation scenarios, reducing repetition and improving visual fidelity. Moreover, UltraImage can generate images up to 6K*6K without low-resolution guidance from a training resolution of 1328p, demonstrating its extreme extrapolation capability. Project page is available at https://thu-ml.github.io/ultraimage.github.io/{https://thu-ml.github.io/ultraimage.github.io/}.

UltraImage：重新思考图像扩散变换器中的分辨率外推

UltraImage: Rethinking Resolution Extrapolation in Image Diffusion Transformers

摘要

Support