Scaling Concept With Text-Guided Diffusion Models
Keywords: Diffusion models, zero-shot applications, image/audio
Abstract: Text-guided diffusion models have revolutionized generative tasks by producing high-fidelity content based on text descriptions. Additionally, they have enabled an editing paradigm where concepts can be replaced through text conditioning. In this work, we explore a novel paradigm: instead of replacing a concept, can we scale it? We conduct an empirical study to investigate concept decomposition trends in text-guided diffusion models. Leveraging these insights, we propose a simple yet effective method, ScalingConcept, designed to enhance or suppress existing concepts in real input without introducing new ones. To systematically evaluate our method, we introduce the WeakConcept-10 dataset. More importantly, ScalingConcept enables a range of novel zero-shot applications across both image and audio domains, including but not limited to canonical pose generation and generative sound highlighting/removal.
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: 8738
Loading