AnyExpress: One Adapter Enabling Highly Flexible Audio-Driven Portrait Animation

Haonan Lin, Kuan Tian, Yan Chen, Feng Tian, Jun Zhang

Keywords: Diffusion-based Portrait Animation, Audio-Driven Talking Face, Plug-and-Play Adapter, Flexible Video Generation

TL;DR: We propose AnyExpress, a versatile and efficient audio-motion adapter that offers unprecedented freedom in generating freeform portrait animations. Its plug-and-play nature allows it to adapt to various base models and control mechanisms.

Abstract: Portrait animation, particularly audio-driven portrait animation, requires flexibility in facial expressions, head movement, and dynamic contexts. However, existing diffusion-based methods rely heavily on the design of ReferenceNet, leading to increased training complexity and incompatibility with other custom base models or adapters, also limiting face position, view changes, and animated context generation. To address these challenges, we propose ***AnyExpress***, a lightweight, modular framework that eliminates the need for ReferenceNet, reducing the number of trainable parameters by **7** times. By training one plug-and-play *audio-motion adapter*, it allows freeform, expressive audio-driven portrait animation with any face pose and any animated context, while supporting text-driven modifications. In the context of character generation, there are two primary methods to control the desired character attributes. First, if a specific ID needs to be assigned, this can be achieved through ID controls (*e.g.*, IP-Adapter-Face). Alternatively, the character’s attributes can be controlled through textual descriptions. Through comprehensive qualitative and quantitative analyses, ***AnyExpress*** demonstrates unprecedented freedom in generating videos with dynamic background, lower training demand, and seamless integration with evolving custom models and control adapters, providing a flexible solution for diverse generation needs. The demo is available at https://anyexpress-alpha.github.io/Any, and we will release our code, encouraging further improvement.

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: 160