English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

We introduce HiDiffusion, a tuning-free framework comprised ofResolution-Aware U-Net (RAU-Net) and Modified Shifted Window Multi-headSelf-Attention (MSW-MSA) to enable pretrained large text-to-image diffusionmodels to efficiently generate high-resolution images (e.g. 1024$\times$1024)that surpass the training image resolution. Pretrained diffusion modelsencounter unreasonable object duplication in generating images beyond thetraining image resolution. We attribute it to the mismatch between the featuremap size of high-resolution images and the receptive field of U-Net'sconvolution. To address this issue, we propose a simple yet scalable methodnamed RAU-Net. RAU-Net dynamically adjusts the feature map size to match theconvolution's receptive field in the deep block of U-Net. Another obstacle inhigh-resolution synthesis is the slow inference speed of U-Net. Ourobservations reveal that the global self-attention in the top block, whichexhibits locality, however, consumes the majority of computational resources.To tackle this issue, we propose MSW-MSA. Unlike previous window attentionmechanisms, our method uses a much larger window size and dynamically shiftswindows to better accommodate diffusion models. Extensive experimentsdemonstrate that our HiDiffusion can scale diffusion models to generate1024$\times$1024, 2048$\times$2048, or even 4096$\times$4096 resolution images,while simultaneously reducing inference time by 40\%-60\%, achievingstate-of-the-art performance on high-resolution image synthesis. The mostsignificant revelation of our work is that a pretrained diffusion model onlow-resolution images is scalable for high-resolution generation withoutfurther tuning. We hope this revelation can provide insights for futureresearch on the scalability of diffusion models.

HiDiffusion: Unlocking High-Resolution Creativity and Efficiency in Low-Resolution Trained Diffusion Models