GuideEdit: Enhancing Face Video Editing with Fine-grained Control
Keywords: Face video editing, Diffusion model
Abstract: Face video editing (FVE) requires maintaining temporal consistency and iden-
tity preservation while manipulating specific attributes. However, existing FVE
methods often introduce unwanted artifacts and affect non-target attributes during
editing. To address these limitations, we propose GuideEdit to enhance the pre-
cision of face video editing. Given the inherent linearity of the latent variables in
the bottleneck layer of the diffusion U-Net model, there exists a linear mapping
between the input and the latent representation. This allows us to extract a latent
basis within the latent space that effectively encodes the key features related to
target facial attributes. By comparing the latent basis of the original video to that
of the manipulated video, we quantify the manipulation degree, which indicates
the extent of changes made. This manipulation degree serves as a guide for deter-
mining the specific components to be edited, then we achieve more precise control
at each denoising step. Integrating this fine-grained control into the editing pro-
cess allows GuideEdit to enhance temporal consistency and preserve identity of
FVE, while minimizing the introduction of artifacts. Extensive experiments on
diverse real-world videos demonstrate the effectiveness of GuideEdit, showcas-
ing its ability to achieve precise, high-quality edits that maintain coherence across
frames and ensure the preservation of essential visual elements.
Supplementary Material: zip
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: 5427
Loading