A Method to 2D Face Generation based on Angle using Condition Diffusion

In this paper, we propose a study on face generation based on Conditional Diffusion. Continuous research is needed to develop face generation models that produce consistent and natural results. In this study, we built a learning network using a Conditional Diffusion-based generative model to generate facial images from multiple angles (0°, 45°, 135°) and expressions such as anger and joy. Experimental results show that the proposed network achieved an FID score of 82.2, while the network using only Transposed Convolution recorded an FID of 121.1. Future research should focus on improving the quality of generated facial images to achieve more natural and consistent results.