Lay-Your-Scene: Open-Vocabulary Text to Layout Generation

Divyansh Srivastava, Xiang Zhang, Chenru Wen, He Wen, Zhuowen Tu

Keywords: layout generation, diffusion models

TL;DR: Open-Vocabulary Natural Scene Layout Generation with Diffusion Transformers

Abstract: We present Lay-Your-Scene (shorthand LayouSyn), a novel diffusion-Transformer based architecture for open-vocabulary natural scene layout generation. Prior works have used close-sourced scene-unaware Large Language models for open-vocabulary layout generation, limiting their widespread use and scene-specific modeling capability. This work presents the first end-to-end text-to-natural-scene-layout generation pipeline that utilizes lightweight open-source language models to predict objects in the scene and a new conditional layout diffusion Transformer trained in a scene-aware manner. Extensive experiments demonstrate that LayouSyn outperforms existing methods on open-vocabulary and closed-vocabulary layout generation and achieves state-of-the-art performance on challenging spatial and numerical reasoning tasks. Additionally, we present two applications of LayouSyn: First, we demonstrate an interesting finding that we can seamlessly combine initialization from the Large Language model to reduce the diffusion sampling steps. Second, we present a new pipeline for adding objects to the image, demonstrating the potential of LayouSyn in image editing applications.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9396