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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Text-to-image (T2I) generation with Stable Diffusion models (SDMs) involveshigh computing demands due to billion-scale parameters. To enhance efficiency,recent studies have reduced sampling steps and applied network quantizationwhile retaining the original architectures. The lack of architectural reductionattempts may stem from worries over expensive retraining for such massivemodels. In this work, we uncover the surprising potential of block pruning andfeature distillation for low-cost general-purpose T2I. By removing severalresidual and attention blocks from the U-Net of SDMs, we achieve 30%~50%reduction in model size, MACs, and latency. We show that distillationretraining is effective even under limited resources: using only 13 A100 daysand a tiny dataset, our compact models can imitate the original SDMs (v1.4 andv2.1-base with over 6,000 A100 days). Benefiting from the transferredknowledge, our BK-SDMs deliver competitive results on zero-shot MS-COCO againstlarger multi-billion parameter models. We further demonstrate the applicabilityof our lightweight backbones in personalized generation and image-to-imagetranslation. Deployment of our models on edge devices attains 4-secondinference. We hope this work can help build small yet powerful diffusion modelswith feasible training budgets. Code and models can be found at:https://github.com/Nota-NetsPresso/BK-SDM

BK-SDM: A Lightweight, Fast, and Cheap Version of Stable Diffusion