Authors:
(1) Yuxuan Yan,Tencent with Equal contributions and [email protected];
(2) Chi Zhang, Tencent with Equal contributions and Corresponding Author, [email protected];
(3) Rui Wang, Tencent and [email protected];
(4) Yichao Zhou, Tencent and [email protected];
(5) Gege Zhang, Tencent and [email protected];
(6) Pei Cheng, Tencent and [email protected];
(7) Bin Fu, Tencent and [email protected];
(8) Gang Yu, Tencentm and [email protected].
3. Method and 3.1. Hybrid Guidance Strategy
3.2. Handling Multiple Identities
4. Experiments
In this section, we present the design and functionalities of our novel framework. Our method fundamentally builds on StableDiffusion [42], with several pivotal modifications, especially in the condition modules catering to hybridguidance image generation. We start by elaborating on our hybrid guidance design in the proposed condition module. Following that, we delve into the mechanism for managing multiple identities within images. Lastly, we discuss the training strategy of our models. The overview of our model structure is shown in Fig. 2.
This paper is available on arxiv under CC0 1.0 DEED license.