Crafting Layered Designs from Pixels
Keywords: Graphic Design, Layered Image Generation
TL;DR: Generate layered designs benefited from non-layered designs.
Abstract: Graphic designs play a vital role in communicating ideas, values, and messages.
During the design process, designers typically organize their work into layers of
text, objects, and backgrounds to facilitate easier editing and customization. However, creating design in such a format requires significant effort and expertise. On
the other hand, with the advancement of GenAI technologies, high quality graphic
designs created in pixel format have become more popular and accessible, while
with the inherent limitation of editability. Despite this limitation, we recognize
the significant reference value of these non-layered designs, as human designers
often derive inspiration from these images to determine layouts or text styles. Motivated by this observation, we propose Accordion, a graphic design generation
framework built around a vision language model playing distinct roles in three
key stages: (1) reference creation, (2) design planning, and (3) layer generation.
By using the reference image as global design guidance, distinct from existing
methods, our approach ensures that elements within the design are visually harmonious. Moreover, through this three-stage framework, Accordion can benefit
from an unlimited supply of AI-generated references. The stage-wise design of
our framework allows for flexible configuration and various applications, such as
starting from user provided references directly with the later two stages. Additionally, it leverages multiple vision experts such as SAM and element removal
models to facilitate the creation of editable graphic layers. Experimental results
show that Accordion generates favourable results on the DesignIntention benchmark, including tasks such as text-to-template, adding text on background, and
text de-rendering. Furthermore, we fully explore the potential of Accordion to
facilitate the creation of design variations, validating its versatility and flexibility
in the whole design workflow.
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: 32
Loading