Face2Diffusion for Fast and Editable Face Personalization

The University of Tokyo  
Conference on Computer Vision and Pattern Recognition (CVPR) 2024
Banner image

Face2Diffusion satisfies challenging text prompts that include multiple conditions while preserving input face identities without individual test-time tuning.


Face personalization aims to insert specific faces, taken from images, into pretrained text-to-image diffusion models. However, it is still challenging for previous methods to preserve both the identity similarity and editability due to overfitting to training samples. In this paper, we propose Face2Diffusion (F2D) for high-editability face personalization. The core idea behind F2D is that removing identity-irrelevant information from the training pipeline prevents the overfitting problem and improves editability of encoded faces. F2D consists of the following three novel components: 1) Multi-scale identity encoder provides well-disentangled identity features while keeping the benefits of multi-scale information, which improves the diversity of camera poses. 2) Expression guidance disentangles face expressions from identities and improves the controllability of face expressions. 3) Classguided denoising regularization encourages models to learn how faces should be denoised, which boosts the textalignment of backgrounds. Extensive experiments on the FaceForensics++ dataset and diverse prompts demonstrate our method greatly improves the trade-off between the identity- and text-fidelity compared to previous state-of-theart methods.


Banner image

(a) During training, we input a face image into our novel multi-scale identity encoder fid and an off-the-shell 3D face reconstruction model fexp to extract identity and expression features, respectively. The concatenated feature is projected into the text space as a word embedding S* by a mapping network fmap. The input image is also encoded by VAE's encoder E and then a Gaussian noise ε is added to it. We constrain the denoised latent feature map to be the original one in the foreground and to be a class-guided denoised result in the background. (b) During inference, the expression feature is replaced with an unconditional vector v˜exp to diversify face expressions of generated images. After injecting the face embedding S* into an input text, the original denoising loop of StableDiffusion is performed to generate an image conditioned by the input face identity and text.

Comparison with Previous Methods

Banner image

Our method generates authentic images aligned with input texts and identities in challenging scenes whereas previous methods compromise an either.

Ablation Study

Banner image

(a) Our MSID encoder disentangles camera poses while keeping the identity similarity. (b) Our expression guidance mitigates overfitting to input face expressions. (c) Our CGDR improves the text-fidelity mainly on backgrounds.


      title={Face2Diffusion for Fast and Editable Face Personalization},
      author={Shiohara, Kaede and Yamasaki, Toshihiko},
      booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},