Stable Diffusion Feature Extraction for Sketching with One Example

Kwan Yun, Youngseo Kim, Kwanggyoon Seo, Chang Wook Seo, Junyong Noh

Keywords: Diffusion Model, Stable Diffusion, Domain Adaptation, Sketch Extraction, Single Shot

TL;DR: Selecting features and controlling conditions of the diffusion model for sketch extraction

Abstract:

Sketching is both a fundamental artistic expression and a crucial aspect of art. The significance of sketching has increased alongside the development of sketch-based generative and editing models.
To enable individuals to use these sketch-based generative models effectively, personalizing sketch extraction is crucial. In response, we introduce $\text{DiffSketch}$, a novel method capable of generating various geometrically aligned sketches from text or images, using a single manual drawing for training the style. Our method exploits rich information available in features from a pretrained Stable Diffusion model to achieve effective domain adaptation. To further streamline the process of sketch extraction, we further refine our approach by distilling the knowledge from the trained generator into the image-to-sketch network, which is termed as $\text{DiffSketch}_{distilled}$. Through a series of comparisons, we verify that our method not only outperforms existing state-of-the-art sketch extraction methods but also surpasses diffusion-based stylization methods in the task of extracting sketches.

Primary Area: applications to computer vision, audio, language, and other modalities

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: 4112